1 /*
2 * Copyright (c) 2003, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Platform-dependent initialization
46
47 void TemplateTable::pd_initialize() {
48 // No amd64 specific initialization
49 }
50
51 // Address computation: local variables
52
53 static inline Address iaddress(int n) {
54 return Address(r14, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 static inline Address faddress(int n) {
62 return iaddress(n);
63 }
64
65 static inline Address daddress(int n) {
66 return laddress(n);
67 }
68
69 static inline Address aaddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address iaddress(Register r) {
74 return Address(r14, r, Address::times_8);
75 }
76
77 static inline Address laddress(Register r) {
78 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
79 }
80
81 static inline Address faddress(Register r) {
82 return iaddress(r);
83 }
84
85 static inline Address daddress(Register r) {
86 return laddress(r);
87 }
88
89 static inline Address aaddress(Register r) {
90 return iaddress(r);
91 }
92
93 static inline Address at_rsp() {
94 return Address(rsp, 0);
95 }
96
97 // At top of Java expression stack which may be different than esp(). It
98 // isn't for category 1 objects.
99 static inline Address at_tos () {
100 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
101 }
102
103 static inline Address at_tos_p1() {
104 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
105 }
106
107 static inline Address at_tos_p2() {
108 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
109 }
110
111 // Condition conversion
112 static Assembler::Condition j_not(TemplateTable::Condition cc) {
113 switch (cc) {
114 case TemplateTable::equal : return Assembler::notEqual;
115 case TemplateTable::not_equal : return Assembler::equal;
116 case TemplateTable::less : return Assembler::greaterEqual;
117 case TemplateTable::less_equal : return Assembler::greater;
118 case TemplateTable::greater : return Assembler::lessEqual;
119 case TemplateTable::greater_equal: return Assembler::less;
120 }
121 ShouldNotReachHere();
122 return Assembler::zero;
123 }
124
125
126 // Miscelaneous helper routines
127 // Store an oop (or NULL) at the address described by obj.
128 // If val == noreg this means store a NULL
129
130 static void do_oop_store(InterpreterMacroAssembler* _masm,
131 Address obj,
132 Register val,
133 BarrierSet::Name barrier,
134 bool precise) {
135 assert(val == noreg || val == rax, "parameter is just for looks");
136 switch (barrier) {
137 #if INCLUDE_ALL_GCS
138 case BarrierSet::G1SATBCT:
139 case BarrierSet::G1SATBCTLogging:
140 {
141 // flatten object address if needed
142 if (obj.index() == noreg && obj.disp() == 0) {
143 if (obj.base() != rdx) {
144 __ movq(rdx, obj.base());
145 }
146 } else {
147 __ leaq(rdx, obj);
148 }
149 __ g1_write_barrier_pre(rdx /* obj */,
150 rbx /* pre_val */,
151 r15_thread /* thread */,
152 r8 /* tmp */,
153 val != noreg /* tosca_live */,
154 false /* expand_call */);
155 if (val == noreg) {
156 __ store_heap_oop_null(Address(rdx, 0));
157 } else {
158 // G1 barrier needs uncompressed oop for region cross check.
159 Register new_val = val;
160 if (UseCompressedOops) {
161 new_val = rbx;
162 __ movptr(new_val, val);
163 }
164 __ store_heap_oop(Address(rdx, 0), val);
165 __ g1_write_barrier_post(rdx /* store_adr */,
166 new_val /* new_val */,
167 r15_thread /* thread */,
168 r8 /* tmp */,
169 rbx /* tmp2 */);
170 }
171 }
172 break;
173 #endif // INCLUDE_ALL_GCS
174 case BarrierSet::CardTableModRef:
175 case BarrierSet::CardTableExtension:
176 {
177 if (val == noreg) {
178 __ store_heap_oop_null(obj);
179 } else {
180 __ store_heap_oop(obj, val);
181 // flatten object address if needed
182 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
183 __ store_check(obj.base());
184 } else {
185 __ leaq(rdx, obj);
186 __ store_check(rdx);
187 }
188 }
189 }
190 break;
191 case BarrierSet::ModRef:
192 case BarrierSet::Other:
193 if (val == noreg) {
194 __ store_heap_oop_null(obj);
195 } else {
196 __ store_heap_oop(obj, val);
197 }
198 break;
199 default :
200 ShouldNotReachHere();
201
202 }
203 }
204
205 Address TemplateTable::at_bcp(int offset) {
206 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
207 return Address(r13, offset);
208 }
209
210 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
211 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
212 int byte_no) {
213 if (!RewriteBytecodes) return;
214 Label L_patch_done;
215
216 switch (bc) {
217 case Bytecodes::_fast_aputfield:
218 case Bytecodes::_fast_bputfield:
219 case Bytecodes::_fast_cputfield:
220 case Bytecodes::_fast_dputfield:
221 case Bytecodes::_fast_fputfield:
222 case Bytecodes::_fast_iputfield:
223 case Bytecodes::_fast_lputfield:
224 case Bytecodes::_fast_sputfield:
225 {
226 // We skip bytecode quickening for putfield instructions when
227 // the put_code written to the constant pool cache is zero.
228 // This is required so that every execution of this instruction
229 // calls out to InterpreterRuntime::resolve_get_put to do
230 // additional, required work.
231 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
232 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
233 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
234 __ movl(bc_reg, bc);
235 __ cmpl(temp_reg, (int) 0);
236 __ jcc(Assembler::zero, L_patch_done); // don't patch
237 }
238 break;
239 default:
240 assert(byte_no == -1, "sanity");
241 // the pair bytecodes have already done the load.
242 if (load_bc_into_bc_reg) {
243 __ movl(bc_reg, bc);
244 }
245 }
246
247 if (JvmtiExport::can_post_breakpoint()) {
248 Label L_fast_patch;
249 // if a breakpoint is present we can't rewrite the stream directly
250 __ movzbl(temp_reg, at_bcp(0));
251 __ cmpl(temp_reg, Bytecodes::_breakpoint);
252 __ jcc(Assembler::notEqual, L_fast_patch);
253 __ get_method(temp_reg);
254 // Let breakpoint table handling rewrite to quicker bytecode
255 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, r13, bc_reg);
256 #ifndef ASSERT
257 __ jmpb(L_patch_done);
258 #else
259 __ jmp(L_patch_done);
260 #endif
261 __ bind(L_fast_patch);
262 }
263
264 #ifdef ASSERT
265 Label L_okay;
266 __ load_unsigned_byte(temp_reg, at_bcp(0));
267 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
268 __ jcc(Assembler::equal, L_okay);
269 __ cmpl(temp_reg, bc_reg);
270 __ jcc(Assembler::equal, L_okay);
271 __ stop("patching the wrong bytecode");
272 __ bind(L_okay);
273 #endif
274
275 // patch bytecode
276 __ movb(at_bcp(0), bc_reg);
277 __ bind(L_patch_done);
278 }
279
280
281 // Individual instructions
282
283 void TemplateTable::nop() {
284 transition(vtos, vtos);
285 // nothing to do
286 }
287
288 void TemplateTable::shouldnotreachhere() {
289 transition(vtos, vtos);
290 __ stop("shouldnotreachhere bytecode");
291 }
292
293 void TemplateTable::aconst_null() {
294 transition(vtos, atos);
295 __ xorl(rax, rax);
296 }
297
298 void TemplateTable::iconst(int value) {
299 transition(vtos, itos);
300 if (value == 0) {
301 __ xorl(rax, rax);
302 } else {
303 __ movl(rax, value);
304 }
305 }
306
307 void TemplateTable::lconst(int value) {
308 transition(vtos, ltos);
309 if (value == 0) {
310 __ xorl(rax, rax);
311 } else {
312 __ movl(rax, value);
313 }
314 }
315
316 void TemplateTable::fconst(int value) {
317 transition(vtos, ftos);
318 static float one = 1.0f, two = 2.0f;
319 switch (value) {
320 case 0:
321 __ xorps(xmm0, xmm0);
322 break;
323 case 1:
324 __ movflt(xmm0, ExternalAddress((address) &one));
325 break;
326 case 2:
327 __ movflt(xmm0, ExternalAddress((address) &two));
328 break;
329 default:
330 ShouldNotReachHere();
331 break;
332 }
333 }
334
335 void TemplateTable::dconst(int value) {
336 transition(vtos, dtos);
337 static double one = 1.0;
338 switch (value) {
339 case 0:
340 __ xorpd(xmm0, xmm0);
341 break;
342 case 1:
343 __ movdbl(xmm0, ExternalAddress((address) &one));
344 break;
345 default:
346 ShouldNotReachHere();
347 break;
348 }
349 }
350
351 void TemplateTable::bipush() {
352 transition(vtos, itos);
353 __ load_signed_byte(rax, at_bcp(1));
354 }
355
356 void TemplateTable::sipush() {
357 transition(vtos, itos);
358 __ load_unsigned_short(rax, at_bcp(1));
359 __ bswapl(rax);
360 __ sarl(rax, 16);
361 }
362
363 void TemplateTable::ldc(bool wide) {
364 transition(vtos, vtos);
365 Label call_ldc, notFloat, notClass, Done;
366
367 if (wide) {
368 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
369 } else {
370 __ load_unsigned_byte(rbx, at_bcp(1));
371 }
372
373 __ get_cpool_and_tags(rcx, rax);
374 const int base_offset = ConstantPool::header_size() * wordSize;
375 const int tags_offset = Array<u1>::base_offset_in_bytes();
376
377 // get type
378 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
379
380 // unresolved class - get the resolved class
381 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
382 __ jccb(Assembler::equal, call_ldc);
383
384 // unresolved class in error state - call into runtime to throw the error
385 // from the first resolution attempt
386 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
387 __ jccb(Assembler::equal, call_ldc);
388
389 // resolved class - need to call vm to get java mirror of the class
390 __ cmpl(rdx, JVM_CONSTANT_Class);
391 __ jcc(Assembler::notEqual, notClass);
392
393 __ bind(call_ldc);
394 __ movl(c_rarg1, wide);
395 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
396 __ push_ptr(rax);
397 __ verify_oop(rax);
398 __ jmp(Done);
399
400 __ bind(notClass);
401 __ cmpl(rdx, JVM_CONSTANT_Float);
402 __ jccb(Assembler::notEqual, notFloat);
403 // ftos
404 __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
405 __ push_f();
406 __ jmp(Done);
407
408 __ bind(notFloat);
409 #ifdef ASSERT
410 {
411 Label L;
412 __ cmpl(rdx, JVM_CONSTANT_Integer);
413 __ jcc(Assembler::equal, L);
414 // String and Object are rewritten to fast_aldc
415 __ stop("unexpected tag type in ldc");
416 __ bind(L);
417 }
418 #endif
419 // itos JVM_CONSTANT_Integer only
420 __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
421 __ push_i(rax);
422 __ bind(Done);
423 }
424
425 // Fast path for caching oop constants.
426 void TemplateTable::fast_aldc(bool wide) {
427 transition(vtos, atos);
428
429 Register result = rax;
430 Register tmp = rdx;
431 int index_size = wide ? sizeof(u2) : sizeof(u1);
432
433 Label resolved;
434
435 // We are resolved if the resolved reference cache entry contains a
436 // non-null object (String, MethodType, etc.)
437 assert_different_registers(result, tmp);
438 __ get_cache_index_at_bcp(tmp, 1, index_size);
439 __ load_resolved_reference_at_index(result, tmp);
440 __ testl(result, result);
441 __ jcc(Assembler::notZero, resolved);
442
443 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
444
445 // first time invocation - must resolve first
446 __ movl(tmp, (int)bytecode());
447 __ call_VM(result, entry, tmp);
448
449 __ bind(resolved);
450
451 if (VerifyOops) {
452 __ verify_oop(result);
453 }
454 }
455
456 void TemplateTable::ldc2_w() {
457 transition(vtos, vtos);
458 Label Long, Done;
459 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
460
461 __ get_cpool_and_tags(rcx, rax);
462 const int base_offset = ConstantPool::header_size() * wordSize;
463 const int tags_offset = Array<u1>::base_offset_in_bytes();
464
465 // get type
466 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
467 JVM_CONSTANT_Double);
468 __ jccb(Assembler::notEqual, Long);
469 // dtos
470 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
471 __ push_d();
472 __ jmpb(Done);
473
474 __ bind(Long);
475 // ltos
476 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
477 __ push_l();
478
479 __ bind(Done);
480 }
481
482 void TemplateTable::locals_index(Register reg, int offset) {
483 __ load_unsigned_byte(reg, at_bcp(offset));
484 __ negptr(reg);
485 }
486
487 void TemplateTable::iload() {
488 transition(vtos, itos);
489 if (RewriteFrequentPairs) {
490 Label rewrite, done;
491 const Register bc = c_rarg3;
492 assert(rbx != bc, "register damaged");
493
494 // get next byte
495 __ load_unsigned_byte(rbx,
496 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
497 // if _iload, wait to rewrite to iload2. We only want to rewrite the
498 // last two iloads in a pair. Comparing against fast_iload means that
499 // the next bytecode is neither an iload or a caload, and therefore
500 // an iload pair.
501 __ cmpl(rbx, Bytecodes::_iload);
502 __ jcc(Assembler::equal, done);
503
504 __ cmpl(rbx, Bytecodes::_fast_iload);
505 __ movl(bc, Bytecodes::_fast_iload2);
506 __ jccb(Assembler::equal, rewrite);
507
508 // if _caload, rewrite to fast_icaload
509 __ cmpl(rbx, Bytecodes::_caload);
510 __ movl(bc, Bytecodes::_fast_icaload);
511 __ jccb(Assembler::equal, rewrite);
512
513 // rewrite so iload doesn't check again.
514 __ movl(bc, Bytecodes::_fast_iload);
515
516 // rewrite
517 // bc: fast bytecode
518 __ bind(rewrite);
519 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
520 __ bind(done);
521 }
522
523 // Get the local value into tos
524 locals_index(rbx);
525 __ movl(rax, iaddress(rbx));
526 }
527
528 void TemplateTable::fast_iload2() {
529 transition(vtos, itos);
530 locals_index(rbx);
531 __ movl(rax, iaddress(rbx));
532 __ push(itos);
533 locals_index(rbx, 3);
534 __ movl(rax, iaddress(rbx));
535 }
536
537 void TemplateTable::fast_iload() {
538 transition(vtos, itos);
539 locals_index(rbx);
540 __ movl(rax, iaddress(rbx));
541 }
542
543 void TemplateTable::lload() {
544 transition(vtos, ltos);
545 locals_index(rbx);
546 __ movq(rax, laddress(rbx));
547 }
548
549 void TemplateTable::fload() {
550 transition(vtos, ftos);
551 locals_index(rbx);
552 __ movflt(xmm0, faddress(rbx));
553 }
554
555 void TemplateTable::dload() {
556 transition(vtos, dtos);
557 locals_index(rbx);
558 __ movdbl(xmm0, daddress(rbx));
559 }
560
561 void TemplateTable::aload() {
562 transition(vtos, atos);
563 locals_index(rbx);
564 __ movptr(rax, aaddress(rbx));
565 }
566
567 void TemplateTable::locals_index_wide(Register reg) {
568 __ load_unsigned_short(reg, at_bcp(2));
569 __ bswapl(reg);
570 __ shrl(reg, 16);
571 __ negptr(reg);
572 }
573
574 void TemplateTable::wide_iload() {
575 transition(vtos, itos);
576 locals_index_wide(rbx);
577 __ movl(rax, iaddress(rbx));
578 }
579
580 void TemplateTable::wide_lload() {
581 transition(vtos, ltos);
582 locals_index_wide(rbx);
583 __ movq(rax, laddress(rbx));
584 }
585
586 void TemplateTable::wide_fload() {
587 transition(vtos, ftos);
588 locals_index_wide(rbx);
589 __ movflt(xmm0, faddress(rbx));
590 }
591
592 void TemplateTable::wide_dload() {
593 transition(vtos, dtos);
594 locals_index_wide(rbx);
595 __ movdbl(xmm0, daddress(rbx));
596 }
597
598 void TemplateTable::wide_aload() {
599 transition(vtos, atos);
600 locals_index_wide(rbx);
601 __ movptr(rax, aaddress(rbx));
602 }
603
604 void TemplateTable::index_check(Register array, Register index) {
605 // destroys rbx
606 // check array
607 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
608 // sign extend index for use by indexed load
609 __ movl2ptr(index, index);
610 // check index
611 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
612 if (index != rbx) {
613 // ??? convention: move aberrant index into ebx for exception message
614 assert(rbx != array, "different registers");
615 __ movl(rbx, index);
616 }
617 __ jump_cc(Assembler::aboveEqual,
618 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
619 }
620
621 void TemplateTable::iaload() {
622 transition(itos, itos);
623 __ pop_ptr(rdx);
624 // eax: index
625 // rdx: array
626 index_check(rdx, rax); // kills rbx
627 __ movl(rax, Address(rdx, rax,
628 Address::times_4,
629 arrayOopDesc::base_offset_in_bytes(T_INT)));
630 }
631
632 void TemplateTable::laload() {
633 transition(itos, ltos);
634 __ pop_ptr(rdx);
635 // eax: index
636 // rdx: array
637 index_check(rdx, rax); // kills rbx
638 __ movq(rax, Address(rdx, rbx,
639 Address::times_8,
640 arrayOopDesc::base_offset_in_bytes(T_LONG)));
641 }
642
643 void TemplateTable::faload() {
644 transition(itos, ftos);
645 __ pop_ptr(rdx);
646 // eax: index
647 // rdx: array
648 index_check(rdx, rax); // kills rbx
649 __ movflt(xmm0, Address(rdx, rax,
650 Address::times_4,
651 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
652 }
653
654 void TemplateTable::daload() {
655 transition(itos, dtos);
656 __ pop_ptr(rdx);
657 // eax: index
658 // rdx: array
659 index_check(rdx, rax); // kills rbx
660 __ movdbl(xmm0, Address(rdx, rax,
661 Address::times_8,
662 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
663 }
664
665 void TemplateTable::aaload() {
666 transition(itos, atos);
667 __ pop_ptr(rdx);
668 // eax: index
669 // rdx: array
670 index_check(rdx, rax); // kills rbx
671 __ load_heap_oop(rax, Address(rdx, rax,
672 UseCompressedOops ? Address::times_4 : Address::times_8,
673 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
674 }
675
676 void TemplateTable::baload() {
677 transition(itos, itos);
678 __ pop_ptr(rdx);
679 // eax: index
680 // rdx: array
681 index_check(rdx, rax); // kills rbx
682 __ load_signed_byte(rax,
683 Address(rdx, rax,
684 Address::times_1,
685 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
686 }
687
688 void TemplateTable::caload() {
689 transition(itos, itos);
690 __ pop_ptr(rdx);
691 // eax: index
692 // rdx: array
693 index_check(rdx, rax); // kills rbx
694 __ load_unsigned_short(rax,
695 Address(rdx, rax,
696 Address::times_2,
697 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
698 }
699
700 // iload followed by caload frequent pair
701 void TemplateTable::fast_icaload() {
702 transition(vtos, itos);
703 // load index out of locals
704 locals_index(rbx);
705 __ movl(rax, iaddress(rbx));
706
707 // eax: index
708 // rdx: array
709 __ pop_ptr(rdx);
710 index_check(rdx, rax); // kills rbx
711 __ load_unsigned_short(rax,
712 Address(rdx, rax,
713 Address::times_2,
714 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
715 }
716
717 void TemplateTable::saload() {
718 transition(itos, itos);
719 __ pop_ptr(rdx);
720 // eax: index
721 // rdx: array
722 index_check(rdx, rax); // kills rbx
723 __ load_signed_short(rax,
724 Address(rdx, rax,
725 Address::times_2,
726 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
727 }
728
729 void TemplateTable::iload(int n) {
730 transition(vtos, itos);
731 __ movl(rax, iaddress(n));
732 }
733
734 void TemplateTable::lload(int n) {
735 transition(vtos, ltos);
736 __ movq(rax, laddress(n));
737 }
738
739 void TemplateTable::fload(int n) {
740 transition(vtos, ftos);
741 __ movflt(xmm0, faddress(n));
742 }
743
744 void TemplateTable::dload(int n) {
745 transition(vtos, dtos);
746 __ movdbl(xmm0, daddress(n));
747 }
748
749 void TemplateTable::aload(int n) {
750 transition(vtos, atos);
751 __ movptr(rax, aaddress(n));
752 }
753
754 void TemplateTable::aload_0() {
755 transition(vtos, atos);
756 // According to bytecode histograms, the pairs:
757 //
758 // _aload_0, _fast_igetfield
759 // _aload_0, _fast_agetfield
760 // _aload_0, _fast_fgetfield
761 //
762 // occur frequently. If RewriteFrequentPairs is set, the (slow)
763 // _aload_0 bytecode checks if the next bytecode is either
764 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
765 // rewrites the current bytecode into a pair bytecode; otherwise it
766 // rewrites the current bytecode into _fast_aload_0 that doesn't do
767 // the pair check anymore.
768 //
769 // Note: If the next bytecode is _getfield, the rewrite must be
770 // delayed, otherwise we may miss an opportunity for a pair.
771 //
772 // Also rewrite frequent pairs
773 // aload_0, aload_1
774 // aload_0, iload_1
775 // These bytecodes with a small amount of code are most profitable
776 // to rewrite
777 if (RewriteFrequentPairs) {
778 Label rewrite, done;
779 const Register bc = c_rarg3;
780 assert(rbx != bc, "register damaged");
781 // get next byte
782 __ load_unsigned_byte(rbx,
783 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
784
785 // do actual aload_0
786 aload(0);
787
788 // if _getfield then wait with rewrite
789 __ cmpl(rbx, Bytecodes::_getfield);
790 __ jcc(Assembler::equal, done);
791
792 // if _igetfield then reqrite to _fast_iaccess_0
793 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
794 Bytecodes::_aload_0,
795 "fix bytecode definition");
796 __ cmpl(rbx, Bytecodes::_fast_igetfield);
797 __ movl(bc, Bytecodes::_fast_iaccess_0);
798 __ jccb(Assembler::equal, rewrite);
799
800 // if _agetfield then reqrite to _fast_aaccess_0
801 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
802 Bytecodes::_aload_0,
803 "fix bytecode definition");
804 __ cmpl(rbx, Bytecodes::_fast_agetfield);
805 __ movl(bc, Bytecodes::_fast_aaccess_0);
806 __ jccb(Assembler::equal, rewrite);
807
808 // if _fgetfield then reqrite to _fast_faccess_0
809 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
810 Bytecodes::_aload_0,
811 "fix bytecode definition");
812 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
813 __ movl(bc, Bytecodes::_fast_faccess_0);
814 __ jccb(Assembler::equal, rewrite);
815
816 // else rewrite to _fast_aload0
817 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
818 Bytecodes::_aload_0,
819 "fix bytecode definition");
820 __ movl(bc, Bytecodes::_fast_aload_0);
821
822 // rewrite
823 // bc: fast bytecode
824 __ bind(rewrite);
825 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
826
827 __ bind(done);
828 } else {
829 aload(0);
830 }
831 }
832
833 void TemplateTable::istore() {
834 transition(itos, vtos);
835 locals_index(rbx);
836 __ movl(iaddress(rbx), rax);
837 }
838
839 void TemplateTable::lstore() {
840 transition(ltos, vtos);
841 locals_index(rbx);
842 __ movq(laddress(rbx), rax);
843 }
844
845 void TemplateTable::fstore() {
846 transition(ftos, vtos);
847 locals_index(rbx);
848 __ movflt(faddress(rbx), xmm0);
849 }
850
851 void TemplateTable::dstore() {
852 transition(dtos, vtos);
853 locals_index(rbx);
854 __ movdbl(daddress(rbx), xmm0);
855 }
856
857 void TemplateTable::astore() {
858 transition(vtos, vtos);
859 __ pop_ptr(rax);
860 locals_index(rbx);
861 __ movptr(aaddress(rbx), rax);
862 }
863
864 void TemplateTable::wide_istore() {
865 transition(vtos, vtos);
866 __ pop_i();
867 locals_index_wide(rbx);
868 __ movl(iaddress(rbx), rax);
869 }
870
871 void TemplateTable::wide_lstore() {
872 transition(vtos, vtos);
873 __ pop_l();
874 locals_index_wide(rbx);
875 __ movq(laddress(rbx), rax);
876 }
877
878 void TemplateTable::wide_fstore() {
879 transition(vtos, vtos);
880 __ pop_f();
881 locals_index_wide(rbx);
882 __ movflt(faddress(rbx), xmm0);
883 }
884
885 void TemplateTable::wide_dstore() {
886 transition(vtos, vtos);
887 __ pop_d();
888 locals_index_wide(rbx);
889 __ movdbl(daddress(rbx), xmm0);
890 }
891
892 void TemplateTable::wide_astore() {
893 transition(vtos, vtos);
894 __ pop_ptr(rax);
895 locals_index_wide(rbx);
896 __ movptr(aaddress(rbx), rax);
897 }
898
899 void TemplateTable::iastore() {
900 transition(itos, vtos);
901 __ pop_i(rbx);
902 __ pop_ptr(rdx);
903 // eax: value
904 // ebx: index
905 // rdx: array
906 index_check(rdx, rbx); // prefer index in ebx
907 __ movl(Address(rdx, rbx,
908 Address::times_4,
909 arrayOopDesc::base_offset_in_bytes(T_INT)),
910 rax);
911 }
912
913 void TemplateTable::lastore() {
914 transition(ltos, vtos);
915 __ pop_i(rbx);
916 __ pop_ptr(rdx);
917 // rax: value
918 // ebx: index
919 // rdx: array
920 index_check(rdx, rbx); // prefer index in ebx
921 __ movq(Address(rdx, rbx,
922 Address::times_8,
923 arrayOopDesc::base_offset_in_bytes(T_LONG)),
924 rax);
925 }
926
927 void TemplateTable::fastore() {
928 transition(ftos, vtos);
929 __ pop_i(rbx);
930 __ pop_ptr(rdx);
931 // xmm0: value
932 // ebx: index
933 // rdx: array
934 index_check(rdx, rbx); // prefer index in ebx
935 __ movflt(Address(rdx, rbx,
936 Address::times_4,
937 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
938 xmm0);
939 }
940
941 void TemplateTable::dastore() {
942 transition(dtos, vtos);
943 __ pop_i(rbx);
944 __ pop_ptr(rdx);
945 // xmm0: value
946 // ebx: index
947 // rdx: array
948 index_check(rdx, rbx); // prefer index in ebx
949 __ movdbl(Address(rdx, rbx,
950 Address::times_8,
951 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
952 xmm0);
953 }
954
955 void TemplateTable::aastore() {
956 Label is_null, ok_is_subtype, done;
957 transition(vtos, vtos);
958 // stack: ..., array, index, value
959 __ movptr(rax, at_tos()); // value
960 __ movl(rcx, at_tos_p1()); // index
961 __ movptr(rdx, at_tos_p2()); // array
962
963 Address element_address(rdx, rcx,
964 UseCompressedOops? Address::times_4 : Address::times_8,
965 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
966
967 index_check(rdx, rcx); // kills rbx
968 // do array store check - check for NULL value first
969 __ testptr(rax, rax);
970 __ jcc(Assembler::zero, is_null);
971
972 // Move subklass into rbx
973 __ load_klass(rbx, rax);
974 // Move superklass into rax
975 __ load_klass(rax, rdx);
976 __ movptr(rax, Address(rax,
977 ObjArrayKlass::element_klass_offset()));
978 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
979 __ lea(rdx, element_address);
980
981 // Generate subtype check. Blows rcx, rdi
982 // Superklass in rax. Subklass in rbx.
983 __ gen_subtype_check(rbx, ok_is_subtype);
984
985 // Come here on failure
986 // object is at TOS
987 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
988
989 // Come here on success
990 __ bind(ok_is_subtype);
991
992 // Get the value we will store
993 __ movptr(rax, at_tos());
994 // Now store using the appropriate barrier
995 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
996 __ jmp(done);
997
998 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
999 __ bind(is_null);
1000 __ profile_null_seen(rbx);
1001
1002 // Store a NULL
1003 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1004
1005 // Pop stack arguments
1006 __ bind(done);
1007 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1008 }
1009
1010 void TemplateTable::bastore() {
1011 transition(itos, vtos);
1012 __ pop_i(rbx);
1013 __ pop_ptr(rdx);
1014 // eax: value
1015 // ebx: index
1016 // rdx: array
1017 index_check(rdx, rbx); // prefer index in ebx
1018 __ movb(Address(rdx, rbx,
1019 Address::times_1,
1020 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1021 rax);
1022 }
1023
1024 void TemplateTable::castore() {
1025 transition(itos, vtos);
1026 __ pop_i(rbx);
1027 __ pop_ptr(rdx);
1028 // eax: value
1029 // ebx: index
1030 // rdx: array
1031 index_check(rdx, rbx); // prefer index in ebx
1032 __ movw(Address(rdx, rbx,
1033 Address::times_2,
1034 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1035 rax);
1036 }
1037
1038 void TemplateTable::sastore() {
1039 castore();
1040 }
1041
1042 void TemplateTable::istore(int n) {
1043 transition(itos, vtos);
1044 __ movl(iaddress(n), rax);
1045 }
1046
1047 void TemplateTable::lstore(int n) {
1048 transition(ltos, vtos);
1049 __ movq(laddress(n), rax);
1050 }
1051
1052 void TemplateTable::fstore(int n) {
1053 transition(ftos, vtos);
1054 __ movflt(faddress(n), xmm0);
1055 }
1056
1057 void TemplateTable::dstore(int n) {
1058 transition(dtos, vtos);
1059 __ movdbl(daddress(n), xmm0);
1060 }
1061
1062 void TemplateTable::astore(int n) {
1063 transition(vtos, vtos);
1064 __ pop_ptr(rax);
1065 __ movptr(aaddress(n), rax);
1066 }
1067
1068 void TemplateTable::pop() {
1069 transition(vtos, vtos);
1070 __ addptr(rsp, Interpreter::stackElementSize);
1071 }
1072
1073 void TemplateTable::pop2() {
1074 transition(vtos, vtos);
1075 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1076 }
1077
1078 void TemplateTable::dup() {
1079 transition(vtos, vtos);
1080 __ load_ptr(0, rax);
1081 __ push_ptr(rax);
1082 // stack: ..., a, a
1083 }
1084
1085 void TemplateTable::dup_x1() {
1086 transition(vtos, vtos);
1087 // stack: ..., a, b
1088 __ load_ptr( 0, rax); // load b
1089 __ load_ptr( 1, rcx); // load a
1090 __ store_ptr(1, rax); // store b
1091 __ store_ptr(0, rcx); // store a
1092 __ push_ptr(rax); // push b
1093 // stack: ..., b, a, b
1094 }
1095
1096 void TemplateTable::dup_x2() {
1097 transition(vtos, vtos);
1098 // stack: ..., a, b, c
1099 __ load_ptr( 0, rax); // load c
1100 __ load_ptr( 2, rcx); // load a
1101 __ store_ptr(2, rax); // store c in a
1102 __ push_ptr(rax); // push c
1103 // stack: ..., c, b, c, c
1104 __ load_ptr( 2, rax); // load b
1105 __ store_ptr(2, rcx); // store a in b
1106 // stack: ..., c, a, c, c
1107 __ store_ptr(1, rax); // store b in c
1108 // stack: ..., c, a, b, c
1109 }
1110
1111 void TemplateTable::dup2() {
1112 transition(vtos, vtos);
1113 // stack: ..., a, b
1114 __ load_ptr(1, rax); // load a
1115 __ push_ptr(rax); // push a
1116 __ load_ptr(1, rax); // load b
1117 __ push_ptr(rax); // push b
1118 // stack: ..., a, b, a, b
1119 }
1120
1121 void TemplateTable::dup2_x1() {
1122 transition(vtos, vtos);
1123 // stack: ..., a, b, c
1124 __ load_ptr( 0, rcx); // load c
1125 __ load_ptr( 1, rax); // load b
1126 __ push_ptr(rax); // push b
1127 __ push_ptr(rcx); // push c
1128 // stack: ..., a, b, c, b, c
1129 __ store_ptr(3, rcx); // store c in b
1130 // stack: ..., a, c, c, b, c
1131 __ load_ptr( 4, rcx); // load a
1132 __ store_ptr(2, rcx); // store a in 2nd c
1133 // stack: ..., a, c, a, b, c
1134 __ store_ptr(4, rax); // store b in a
1135 // stack: ..., b, c, a, b, c
1136 }
1137
1138 void TemplateTable::dup2_x2() {
1139 transition(vtos, vtos);
1140 // stack: ..., a, b, c, d
1141 __ load_ptr( 0, rcx); // load d
1142 __ load_ptr( 1, rax); // load c
1143 __ push_ptr(rax); // push c
1144 __ push_ptr(rcx); // push d
1145 // stack: ..., a, b, c, d, c, d
1146 __ load_ptr( 4, rax); // load b
1147 __ store_ptr(2, rax); // store b in d
1148 __ store_ptr(4, rcx); // store d in b
1149 // stack: ..., a, d, c, b, c, d
1150 __ load_ptr( 5, rcx); // load a
1151 __ load_ptr( 3, rax); // load c
1152 __ store_ptr(3, rcx); // store a in c
1153 __ store_ptr(5, rax); // store c in a
1154 // stack: ..., c, d, a, b, c, d
1155 }
1156
1157 void TemplateTable::swap() {
1158 transition(vtos, vtos);
1159 // stack: ..., a, b
1160 __ load_ptr( 1, rcx); // load a
1161 __ load_ptr( 0, rax); // load b
1162 __ store_ptr(0, rcx); // store a in b
1163 __ store_ptr(1, rax); // store b in a
1164 // stack: ..., b, a
1165 }
1166
1167 void TemplateTable::iop2(Operation op) {
1168 transition(itos, itos);
1169 switch (op) {
1170 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1171 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1172 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1173 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1174 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1175 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1176 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1177 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1178 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1179 default : ShouldNotReachHere();
1180 }
1181 }
1182
1183 void TemplateTable::lop2(Operation op) {
1184 transition(ltos, ltos);
1185 switch (op) {
1186 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1187 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1188 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1189 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1190 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1191 default : ShouldNotReachHere();
1192 }
1193 }
1194
1195 void TemplateTable::idiv() {
1196 transition(itos, itos);
1197 __ movl(rcx, rax);
1198 __ pop_i(rax);
1199 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1200 // they are not equal, one could do a normal division (no correction
1201 // needed), which may speed up this implementation for the common case.
1202 // (see also JVM spec., p.243 & p.271)
1203 __ corrected_idivl(rcx);
1204 }
1205
1206 void TemplateTable::irem() {
1207 transition(itos, itos);
1208 __ movl(rcx, rax);
1209 __ pop_i(rax);
1210 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1211 // they are not equal, one could do a normal division (no correction
1212 // needed), which may speed up this implementation for the common case.
1213 // (see also JVM spec., p.243 & p.271)
1214 __ corrected_idivl(rcx);
1215 __ movl(rax, rdx);
1216 }
1217
1218 void TemplateTable::lmul() {
1219 transition(ltos, ltos);
1220 __ pop_l(rdx);
1221 __ imulq(rax, rdx);
1222 }
1223
1224 void TemplateTable::ldiv() {
1225 transition(ltos, ltos);
1226 __ mov(rcx, rax);
1227 __ pop_l(rax);
1228 // generate explicit div0 check
1229 __ testq(rcx, rcx);
1230 __ jump_cc(Assembler::zero,
1231 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1232 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1233 // they are not equal, one could do a normal division (no correction
1234 // needed), which may speed up this implementation for the common case.
1235 // (see also JVM spec., p.243 & p.271)
1236 __ corrected_idivq(rcx); // kills rbx
1237 }
1238
1239 void TemplateTable::lrem() {
1240 transition(ltos, ltos);
1241 __ mov(rcx, rax);
1242 __ pop_l(rax);
1243 __ testq(rcx, rcx);
1244 __ jump_cc(Assembler::zero,
1245 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1246 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1247 // they are not equal, one could do a normal division (no correction
1248 // needed), which may speed up this implementation for the common case.
1249 // (see also JVM spec., p.243 & p.271)
1250 __ corrected_idivq(rcx); // kills rbx
1251 __ mov(rax, rdx);
1252 }
1253
1254 void TemplateTable::lshl() {
1255 transition(itos, ltos);
1256 __ movl(rcx, rax); // get shift count
1257 __ pop_l(rax); // get shift value
1258 __ shlq(rax);
1259 }
1260
1261 void TemplateTable::lshr() {
1262 transition(itos, ltos);
1263 __ movl(rcx, rax); // get shift count
1264 __ pop_l(rax); // get shift value
1265 __ sarq(rax);
1266 }
1267
1268 void TemplateTable::lushr() {
1269 transition(itos, ltos);
1270 __ movl(rcx, rax); // get shift count
1271 __ pop_l(rax); // get shift value
1272 __ shrq(rax);
1273 }
1274
1275 void TemplateTable::fop2(Operation op) {
1276 transition(ftos, ftos);
1277 switch (op) {
1278 case add:
1279 __ addss(xmm0, at_rsp());
1280 __ addptr(rsp, Interpreter::stackElementSize);
1281 break;
1282 case sub:
1283 __ movflt(xmm1, xmm0);
1284 __ pop_f(xmm0);
1285 __ subss(xmm0, xmm1);
1286 break;
1287 case mul:
1288 __ mulss(xmm0, at_rsp());
1289 __ addptr(rsp, Interpreter::stackElementSize);
1290 break;
1291 case div:
1292 __ movflt(xmm1, xmm0);
1293 __ pop_f(xmm0);
1294 __ divss(xmm0, xmm1);
1295 break;
1296 case rem:
1297 __ movflt(xmm1, xmm0);
1298 __ pop_f(xmm0);
1299 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1300 break;
1301 default:
1302 ShouldNotReachHere();
1303 break;
1304 }
1305 }
1306
1307 void TemplateTable::dop2(Operation op) {
1308 transition(dtos, dtos);
1309 switch (op) {
1310 case add:
1311 __ addsd(xmm0, at_rsp());
1312 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1313 break;
1314 case sub:
1315 __ movdbl(xmm1, xmm0);
1316 __ pop_d(xmm0);
1317 __ subsd(xmm0, xmm1);
1318 break;
1319 case mul:
1320 __ mulsd(xmm0, at_rsp());
1321 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1322 break;
1323 case div:
1324 __ movdbl(xmm1, xmm0);
1325 __ pop_d(xmm0);
1326 __ divsd(xmm0, xmm1);
1327 break;
1328 case rem:
1329 __ movdbl(xmm1, xmm0);
1330 __ pop_d(xmm0);
1331 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1332 break;
1333 default:
1334 ShouldNotReachHere();
1335 break;
1336 }
1337 }
1338
1339 void TemplateTable::ineg() {
1340 transition(itos, itos);
1341 __ negl(rax);
1342 }
1343
1344 void TemplateTable::lneg() {
1345 transition(ltos, ltos);
1346 __ negq(rax);
1347 }
1348
1349 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1350 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1351 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1352 // of 128-bits operands for SSE instructions.
1353 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1354 // Store the value to a 128-bits operand.
1355 operand[0] = lo;
1356 operand[1] = hi;
1357 return operand;
1358 }
1359
1360 // Buffer for 128-bits masks used by SSE instructions.
1361 static jlong float_signflip_pool[2*2];
1362 static jlong double_signflip_pool[2*2];
1363
1364 void TemplateTable::fneg() {
1365 transition(ftos, ftos);
1366 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1367 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1368 }
1369
1370 void TemplateTable::dneg() {
1371 transition(dtos, dtos);
1372 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1373 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1374 }
1375
1376 void TemplateTable::iinc() {
1377 transition(vtos, vtos);
1378 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1379 locals_index(rbx);
1380 __ addl(iaddress(rbx), rdx);
1381 }
1382
1383 void TemplateTable::wide_iinc() {
1384 transition(vtos, vtos);
1385 __ movl(rdx, at_bcp(4)); // get constant
1386 locals_index_wide(rbx);
1387 __ bswapl(rdx); // swap bytes & sign-extend constant
1388 __ sarl(rdx, 16);
1389 __ addl(iaddress(rbx), rdx);
1390 // Note: should probably use only one movl to get both
1391 // the index and the constant -> fix this
1392 }
1393
1394 void TemplateTable::convert() {
1395 // Checking
1396 #ifdef ASSERT
1397 {
1398 TosState tos_in = ilgl;
1399 TosState tos_out = ilgl;
1400 switch (bytecode()) {
1401 case Bytecodes::_i2l: // fall through
1402 case Bytecodes::_i2f: // fall through
1403 case Bytecodes::_i2d: // fall through
1404 case Bytecodes::_i2b: // fall through
1405 case Bytecodes::_i2c: // fall through
1406 case Bytecodes::_i2s: tos_in = itos; break;
1407 case Bytecodes::_l2i: // fall through
1408 case Bytecodes::_l2f: // fall through
1409 case Bytecodes::_l2d: tos_in = ltos; break;
1410 case Bytecodes::_f2i: // fall through
1411 case Bytecodes::_f2l: // fall through
1412 case Bytecodes::_f2d: tos_in = ftos; break;
1413 case Bytecodes::_d2i: // fall through
1414 case Bytecodes::_d2l: // fall through
1415 case Bytecodes::_d2f: tos_in = dtos; break;
1416 default : ShouldNotReachHere();
1417 }
1418 switch (bytecode()) {
1419 case Bytecodes::_l2i: // fall through
1420 case Bytecodes::_f2i: // fall through
1421 case Bytecodes::_d2i: // fall through
1422 case Bytecodes::_i2b: // fall through
1423 case Bytecodes::_i2c: // fall through
1424 case Bytecodes::_i2s: tos_out = itos; break;
1425 case Bytecodes::_i2l: // fall through
1426 case Bytecodes::_f2l: // fall through
1427 case Bytecodes::_d2l: tos_out = ltos; break;
1428 case Bytecodes::_i2f: // fall through
1429 case Bytecodes::_l2f: // fall through
1430 case Bytecodes::_d2f: tos_out = ftos; break;
1431 case Bytecodes::_i2d: // fall through
1432 case Bytecodes::_l2d: // fall through
1433 case Bytecodes::_f2d: tos_out = dtos; break;
1434 default : ShouldNotReachHere();
1435 }
1436 transition(tos_in, tos_out);
1437 }
1438 #endif // ASSERT
1439
1440 static const int64_t is_nan = 0x8000000000000000L;
1441
1442 // Conversion
1443 switch (bytecode()) {
1444 case Bytecodes::_i2l:
1445 __ movslq(rax, rax);
1446 break;
1447 case Bytecodes::_i2f:
1448 __ cvtsi2ssl(xmm0, rax);
1449 break;
1450 case Bytecodes::_i2d:
1451 __ cvtsi2sdl(xmm0, rax);
1452 break;
1453 case Bytecodes::_i2b:
1454 __ movsbl(rax, rax);
1455 break;
1456 case Bytecodes::_i2c:
1457 __ movzwl(rax, rax);
1458 break;
1459 case Bytecodes::_i2s:
1460 __ movswl(rax, rax);
1461 break;
1462 case Bytecodes::_l2i:
1463 __ movl(rax, rax);
1464 break;
1465 case Bytecodes::_l2f:
1466 __ cvtsi2ssq(xmm0, rax);
1467 break;
1468 case Bytecodes::_l2d:
1469 __ cvtsi2sdq(xmm0, rax);
1470 break;
1471 case Bytecodes::_f2i:
1472 {
1473 Label L;
1474 __ cvttss2sil(rax, xmm0);
1475 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1476 __ jcc(Assembler::notEqual, L);
1477 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1478 __ bind(L);
1479 }
1480 break;
1481 case Bytecodes::_f2l:
1482 {
1483 Label L;
1484 __ cvttss2siq(rax, xmm0);
1485 // NaN or overflow/underflow?
1486 __ cmp64(rax, ExternalAddress((address) &is_nan));
1487 __ jcc(Assembler::notEqual, L);
1488 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1489 __ bind(L);
1490 }
1491 break;
1492 case Bytecodes::_f2d:
1493 __ cvtss2sd(xmm0, xmm0);
1494 break;
1495 case Bytecodes::_d2i:
1496 {
1497 Label L;
1498 __ cvttsd2sil(rax, xmm0);
1499 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1500 __ jcc(Assembler::notEqual, L);
1501 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1502 __ bind(L);
1503 }
1504 break;
1505 case Bytecodes::_d2l:
1506 {
1507 Label L;
1508 __ cvttsd2siq(rax, xmm0);
1509 // NaN or overflow/underflow?
1510 __ cmp64(rax, ExternalAddress((address) &is_nan));
1511 __ jcc(Assembler::notEqual, L);
1512 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1513 __ bind(L);
1514 }
1515 break;
1516 case Bytecodes::_d2f:
1517 __ cvtsd2ss(xmm0, xmm0);
1518 break;
1519 default:
1520 ShouldNotReachHere();
1521 }
1522 }
1523
1524 void TemplateTable::lcmp() {
1525 transition(ltos, itos);
1526 Label done;
1527 __ pop_l(rdx);
1528 __ cmpq(rdx, rax);
1529 __ movl(rax, -1);
1530 __ jccb(Assembler::less, done);
1531 __ setb(Assembler::notEqual, rax);
1532 __ movzbl(rax, rax);
1533 __ bind(done);
1534 }
1535
1536 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1537 Label done;
1538 if (is_float) {
1539 // XXX get rid of pop here, use ... reg, mem32
1540 __ pop_f(xmm1);
1541 __ ucomiss(xmm1, xmm0);
1542 } else {
1543 // XXX get rid of pop here, use ... reg, mem64
1544 __ pop_d(xmm1);
1545 __ ucomisd(xmm1, xmm0);
1546 }
1547 if (unordered_result < 0) {
1548 __ movl(rax, -1);
1549 __ jccb(Assembler::parity, done);
1550 __ jccb(Assembler::below, done);
1551 __ setb(Assembler::notEqual, rdx);
1552 __ movzbl(rax, rdx);
1553 } else {
1554 __ movl(rax, 1);
1555 __ jccb(Assembler::parity, done);
1556 __ jccb(Assembler::above, done);
1557 __ movl(rax, 0);
1558 __ jccb(Assembler::equal, done);
1559 __ decrementl(rax);
1560 }
1561 __ bind(done);
1562 }
1563
1564 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1565 __ get_method(rcx); // rcx holds method
1566 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1567 // holds bumped taken count
1568
1569 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1570 InvocationCounter::counter_offset();
1571 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1572 InvocationCounter::counter_offset();
1573
1574 // Load up edx with the branch displacement
1575 if (is_wide) {
1576 __ movl(rdx, at_bcp(1));
1577 } else {
1578 __ load_signed_short(rdx, at_bcp(1));
1579 }
1580 __ bswapl(rdx);
1581
1582 if (!is_wide) {
1583 __ sarl(rdx, 16);
1584 }
1585 __ movl2ptr(rdx, rdx);
1586
1587 // Handle all the JSR stuff here, then exit.
1588 // It's much shorter and cleaner than intermingling with the non-JSR
1589 // normal-branch stuff occurring below.
1590 if (is_jsr) {
1591 // Pre-load the next target bytecode into rbx
1592 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1593
1594 // compute return address as bci in rax
1595 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1596 in_bytes(ConstMethod::codes_offset())));
1597 __ subptr(rax, Address(rcx, Method::const_offset()));
1598 // Adjust the bcp in r13 by the displacement in rdx
1599 __ addptr(r13, rdx);
1600 // jsr returns atos that is not an oop
1601 __ push_i(rax);
1602 __ dispatch_only(vtos);
1603 return;
1604 }
1605
1606 // Normal (non-jsr) branch handling
1607
1608 // Adjust the bcp in r13 by the displacement in rdx
1609 __ addptr(r13, rdx);
1610
1611 assert(UseLoopCounter || !UseOnStackReplacement,
1612 "on-stack-replacement requires loop counters");
1613 Label backedge_counter_overflow;
1614 Label profile_method;
1615 Label dispatch;
1616 if (UseLoopCounter) {
1617 // increment backedge counter for backward branches
1618 // rax: MDO
1619 // ebx: MDO bumped taken-count
1620 // rcx: method
1621 // rdx: target offset
1622 // r13: target bcp
1623 // r14: locals pointer
1624 __ testl(rdx, rdx); // check if forward or backward branch
1625 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1626
1627 // check if MethodCounters exists
1628 Label has_counters;
1629 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1630 __ testptr(rax, rax);
1631 __ jcc(Assembler::notZero, has_counters);
1632 __ push(rdx);
1633 __ push(rcx);
1634 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1635 rcx);
1636 __ pop(rcx);
1637 __ pop(rdx);
1638 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1639 __ jcc(Assembler::zero, dispatch);
1640 __ bind(has_counters);
1641
1642 if (TieredCompilation) {
1643 Label no_mdo;
1644 int increment = InvocationCounter::count_increment;
1645 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1646 if (ProfileInterpreter) {
1647 // Are we profiling?
1648 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
1649 __ testptr(rbx, rbx);
1650 __ jccb(Assembler::zero, no_mdo);
1651 // Increment the MDO backedge counter
1652 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
1653 in_bytes(InvocationCounter::counter_offset()));
1654 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1655 rax, false, Assembler::zero, &backedge_counter_overflow);
1656 __ jmp(dispatch);
1657 }
1658 __ bind(no_mdo);
1659 // Increment backedge counter in MethodCounters*
1660 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1661 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1662 rax, false, Assembler::zero, &backedge_counter_overflow);
1663 } else {
1664 // increment counter
1665 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1666 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1667 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1668 __ movl(Address(rcx, be_offset), rax); // store counter
1669
1670 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1671
1672 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1673 __ addl(rax, Address(rcx, be_offset)); // add both counters
1674
1675 if (ProfileInterpreter) {
1676 // Test to see if we should create a method data oop
1677 __ cmp32(rax,
1678 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1679 __ jcc(Assembler::less, dispatch);
1680
1681 // if no method data exists, go to profile method
1682 __ test_method_data_pointer(rax, profile_method);
1683
1684 if (UseOnStackReplacement) {
1685 // check for overflow against ebx which is the MDO taken count
1686 __ cmp32(rbx,
1687 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1688 __ jcc(Assembler::below, dispatch);
1689
1690 // When ProfileInterpreter is on, the backedge_count comes
1691 // from the MethodData*, which value does not get reset on
1692 // the call to frequency_counter_overflow(). To avoid
1693 // excessive calls to the overflow routine while the method is
1694 // being compiled, add a second test to make sure the overflow
1695 // function is called only once every overflow_frequency.
1696 const int overflow_frequency = 1024;
1697 __ andl(rbx, overflow_frequency - 1);
1698 __ jcc(Assembler::zero, backedge_counter_overflow);
1699
1700 }
1701 } else {
1702 if (UseOnStackReplacement) {
1703 // check for overflow against eax, which is the sum of the
1704 // counters
1705 __ cmp32(rax,
1706 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1707 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1708
1709 }
1710 }
1711 }
1712 __ bind(dispatch);
1713 }
1714
1715 // Pre-load the next target bytecode into rbx
1716 __ load_unsigned_byte(rbx, Address(r13, 0));
1717
1718 // continue with the bytecode @ target
1719 // eax: return bci for jsr's, unused otherwise
1720 // ebx: target bytecode
1721 // r13: target bcp
1722 __ dispatch_only(vtos);
1723
1724 if (UseLoopCounter) {
1725 if (ProfileInterpreter) {
1726 // Out-of-line code to allocate method data oop.
1727 __ bind(profile_method);
1728 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1729 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1730 __ set_method_data_pointer_for_bcp();
1731 __ jmp(dispatch);
1732 }
1733
1734 if (UseOnStackReplacement) {
1735 // invocation counter overflow
1736 __ bind(backedge_counter_overflow);
1737 __ negptr(rdx);
1738 __ addptr(rdx, r13); // branch bcp
1739 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1740 __ call_VM(noreg,
1741 CAST_FROM_FN_PTR(address,
1742 InterpreterRuntime::frequency_counter_overflow),
1743 rdx);
1744 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1745
1746 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1747 // ebx: target bytecode
1748 // rdx: scratch
1749 // r14: locals pointer
1750 // r13: bcp
1751 __ testptr(rax, rax); // test result
1752 __ jcc(Assembler::zero, dispatch); // no osr if null
1753 // nmethod may have been invalidated (VM may block upon call_VM return)
1754 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
1755 __ jcc(Assembler::notEqual, dispatch);
1756
1757 // We have the address of an on stack replacement routine in eax
1758 // We need to prepare to execute the OSR method. First we must
1759 // migrate the locals and monitors off of the stack.
1760
1761 __ mov(r13, rax); // save the nmethod
1762
1763 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1764
1765 // eax is OSR buffer, move it to expected parameter location
1766 __ mov(j_rarg0, rax);
1767
1768 // We use j_rarg definitions here so that registers don't conflict as parameter
1769 // registers change across platforms as we are in the midst of a calling
1770 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1771
1772 const Register retaddr = j_rarg2;
1773 const Register sender_sp = j_rarg1;
1774
1775 // pop the interpreter frame
1776 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1777 __ leave(); // remove frame anchor
1778 __ pop(retaddr); // get return address
1779 __ mov(rsp, sender_sp); // set sp to sender sp
1780 // Ensure compiled code always sees stack at proper alignment
1781 __ andptr(rsp, -(StackAlignmentInBytes));
1782
1783 // unlike x86 we need no specialized return from compiled code
1784 // to the interpreter or the call stub.
1785
1786 // push the return address
1787 __ push(retaddr);
1788
1789 // and begin the OSR nmethod
1790 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1791 }
1792 }
1793 }
1794
1795
1796 void TemplateTable::if_0cmp(Condition cc) {
1797 transition(itos, vtos);
1798 // assume branch is more often taken than not (loops use backward branches)
1799 Label not_taken;
1800 __ testl(rax, rax);
1801 __ jcc(j_not(cc), not_taken);
1802 branch(false, false);
1803 __ bind(not_taken);
1804 __ profile_not_taken_branch(rax);
1805 }
1806
1807 void TemplateTable::if_icmp(Condition cc) {
1808 transition(itos, vtos);
1809 // assume branch is more often taken than not (loops use backward branches)
1810 Label not_taken;
1811 __ pop_i(rdx);
1812 __ cmpl(rdx, rax);
1813 __ jcc(j_not(cc), not_taken);
1814 branch(false, false);
1815 __ bind(not_taken);
1816 __ profile_not_taken_branch(rax);
1817 }
1818
1819 void TemplateTable::if_nullcmp(Condition cc) {
1820 transition(atos, vtos);
1821 // assume branch is more often taken than not (loops use backward branches)
1822 Label not_taken;
1823 __ testptr(rax, rax);
1824 __ jcc(j_not(cc), not_taken);
1825 branch(false, false);
1826 __ bind(not_taken);
1827 __ profile_not_taken_branch(rax);
1828 }
1829
1830 void TemplateTable::if_acmp(Condition cc) {
1831 transition(atos, vtos);
1832 // assume branch is more often taken than not (loops use backward branches)
1833 Label not_taken;
1834 __ pop_ptr(rdx);
1835 __ cmpptr(rdx, rax);
1836 __ jcc(j_not(cc), not_taken);
1837 branch(false, false);
1838 __ bind(not_taken);
1839 __ profile_not_taken_branch(rax);
1840 }
1841
1842 void TemplateTable::ret() {
1843 transition(vtos, vtos);
1844 locals_index(rbx);
1845 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1846 __ profile_ret(rbx, rcx);
1847 __ get_method(rax);
1848 __ movptr(r13, Address(rax, Method::const_offset()));
1849 __ lea(r13, Address(r13, rbx, Address::times_1,
1850 ConstMethod::codes_offset()));
1851 __ dispatch_next(vtos);
1852 }
1853
1854 void TemplateTable::wide_ret() {
1855 transition(vtos, vtos);
1856 locals_index_wide(rbx);
1857 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1858 __ profile_ret(rbx, rcx);
1859 __ get_method(rax);
1860 __ movptr(r13, Address(rax, Method::const_offset()));
1861 __ lea(r13, Address(r13, rbx, Address::times_1, ConstMethod::codes_offset()));
1862 __ dispatch_next(vtos);
1863 }
1864
1865 void TemplateTable::tableswitch() {
1866 Label default_case, continue_execution;
1867 transition(itos, vtos);
1868 // align r13
1869 __ lea(rbx, at_bcp(BytesPerInt));
1870 __ andptr(rbx, -BytesPerInt);
1871 // load lo & hi
1872 __ movl(rcx, Address(rbx, BytesPerInt));
1873 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1874 __ bswapl(rcx);
1875 __ bswapl(rdx);
1876 // check against lo & hi
1877 __ cmpl(rax, rcx);
1878 __ jcc(Assembler::less, default_case);
1879 __ cmpl(rax, rdx);
1880 __ jcc(Assembler::greater, default_case);
1881 // lookup dispatch offset
1882 __ subl(rax, rcx);
1883 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1884 __ profile_switch_case(rax, rbx, rcx);
1885 // continue execution
1886 __ bind(continue_execution);
1887 __ bswapl(rdx);
1888 __ movl2ptr(rdx, rdx);
1889 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1890 __ addptr(r13, rdx);
1891 __ dispatch_only(vtos);
1892 // handle default
1893 __ bind(default_case);
1894 __ profile_switch_default(rax);
1895 __ movl(rdx, Address(rbx, 0));
1896 __ jmp(continue_execution);
1897 }
1898
1899 void TemplateTable::lookupswitch() {
1900 transition(itos, itos);
1901 __ stop("lookupswitch bytecode should have been rewritten");
1902 }
1903
1904 void TemplateTable::fast_linearswitch() {
1905 transition(itos, vtos);
1906 Label loop_entry, loop, found, continue_execution;
1907 // bswap rax so we can avoid bswapping the table entries
1908 __ bswapl(rax);
1909 // align r13
1910 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1911 // this instruction (change offsets
1912 // below)
1913 __ andptr(rbx, -BytesPerInt);
1914 // set counter
1915 __ movl(rcx, Address(rbx, BytesPerInt));
1916 __ bswapl(rcx);
1917 __ jmpb(loop_entry);
1918 // table search
1919 __ bind(loop);
1920 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1921 __ jcc(Assembler::equal, found);
1922 __ bind(loop_entry);
1923 __ decrementl(rcx);
1924 __ jcc(Assembler::greaterEqual, loop);
1925 // default case
1926 __ profile_switch_default(rax);
1927 __ movl(rdx, Address(rbx, 0));
1928 __ jmp(continue_execution);
1929 // entry found -> get offset
1930 __ bind(found);
1931 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1932 __ profile_switch_case(rcx, rax, rbx);
1933 // continue execution
1934 __ bind(continue_execution);
1935 __ bswapl(rdx);
1936 __ movl2ptr(rdx, rdx);
1937 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1938 __ addptr(r13, rdx);
1939 __ dispatch_only(vtos);
1940 }
1941
1942 void TemplateTable::fast_binaryswitch() {
1943 transition(itos, vtos);
1944 // Implementation using the following core algorithm:
1945 //
1946 // int binary_search(int key, LookupswitchPair* array, int n) {
1947 // // Binary search according to "Methodik des Programmierens" by
1948 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1949 // int i = 0;
1950 // int j = n;
1951 // while (i+1 < j) {
1952 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1953 // // with Q: for all i: 0 <= i < n: key < a[i]
1954 // // where a stands for the array and assuming that the (inexisting)
1955 // // element a[n] is infinitely big.
1956 // int h = (i + j) >> 1;
1957 // // i < h < j
1958 // if (key < array[h].fast_match()) {
1959 // j = h;
1960 // } else {
1961 // i = h;
1962 // }
1963 // }
1964 // // R: a[i] <= key < a[i+1] or Q
1965 // // (i.e., if key is within array, i is the correct index)
1966 // return i;
1967 // }
1968
1969 // Register allocation
1970 const Register key = rax; // already set (tosca)
1971 const Register array = rbx;
1972 const Register i = rcx;
1973 const Register j = rdx;
1974 const Register h = rdi;
1975 const Register temp = rsi;
1976
1977 // Find array start
1978 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1979 // get rid of this
1980 // instruction (change
1981 // offsets below)
1982 __ andptr(array, -BytesPerInt);
1983
1984 // Initialize i & j
1985 __ xorl(i, i); // i = 0;
1986 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1987
1988 // Convert j into native byteordering
1989 __ bswapl(j);
1990
1991 // And start
1992 Label entry;
1993 __ jmp(entry);
1994
1995 // binary search loop
1996 {
1997 Label loop;
1998 __ bind(loop);
1999 // int h = (i + j) >> 1;
2000 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2001 __ sarl(h, 1); // h = (i + j) >> 1;
2002 // if (key < array[h].fast_match()) {
2003 // j = h;
2004 // } else {
2005 // i = h;
2006 // }
2007 // Convert array[h].match to native byte-ordering before compare
2008 __ movl(temp, Address(array, h, Address::times_8));
2009 __ bswapl(temp);
2010 __ cmpl(key, temp);
2011 // j = h if (key < array[h].fast_match())
2012 __ cmovl(Assembler::less, j, h);
2013 // i = h if (key >= array[h].fast_match())
2014 __ cmovl(Assembler::greaterEqual, i, h);
2015 // while (i+1 < j)
2016 __ bind(entry);
2017 __ leal(h, Address(i, 1)); // i+1
2018 __ cmpl(h, j); // i+1 < j
2019 __ jcc(Assembler::less, loop);
2020 }
2021
2022 // end of binary search, result index is i (must check again!)
2023 Label default_case;
2024 // Convert array[i].match to native byte-ordering before compare
2025 __ movl(temp, Address(array, i, Address::times_8));
2026 __ bswapl(temp);
2027 __ cmpl(key, temp);
2028 __ jcc(Assembler::notEqual, default_case);
2029
2030 // entry found -> j = offset
2031 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2032 __ profile_switch_case(i, key, array);
2033 __ bswapl(j);
2034 __ movl2ptr(j, j);
2035 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2036 __ addptr(r13, j);
2037 __ dispatch_only(vtos);
2038
2039 // default case -> j = default offset
2040 __ bind(default_case);
2041 __ profile_switch_default(i);
2042 __ movl(j, Address(array, -2 * BytesPerInt));
2043 __ bswapl(j);
2044 __ movl2ptr(j, j);
2045 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2046 __ addptr(r13, j);
2047 __ dispatch_only(vtos);
2048 }
2049
2050
2051 void TemplateTable::_return(TosState state) {
2052 transition(state, state);
2053 assert(_desc->calls_vm(),
2054 "inconsistent calls_vm information"); // call in remove_activation
2055
2056 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2057 assert(state == vtos, "only valid state");
2058 __ movptr(c_rarg1, aaddress(0));
2059 __ load_klass(rdi, c_rarg1);
2060 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2061 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2062 Label skip_register_finalizer;
2063 __ jcc(Assembler::zero, skip_register_finalizer);
2064
2065 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2066
2067 __ bind(skip_register_finalizer);
2068 }
2069
2070 __ remove_activation(state, r13);
2071 __ jmp(r13);
2072 }
2073
2074 // ----------------------------------------------------------------------------
2075 // Volatile variables demand their effects be made known to all CPU's
2076 // in order. Store buffers on most chips allow reads & writes to
2077 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2078 // without some kind of memory barrier (i.e., it's not sufficient that
2079 // the interpreter does not reorder volatile references, the hardware
2080 // also must not reorder them).
2081 //
2082 // According to the new Java Memory Model (JMM):
2083 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2084 // writes act as aquire & release, so:
2085 // (2) A read cannot let unrelated NON-volatile memory refs that
2086 // happen after the read float up to before the read. It's OK for
2087 // non-volatile memory refs that happen before the volatile read to
2088 // float down below it.
2089 // (3) Similar a volatile write cannot let unrelated NON-volatile
2090 // memory refs that happen BEFORE the write float down to after the
2091 // write. It's OK for non-volatile memory refs that happen after the
2092 // volatile write to float up before it.
2093 //
2094 // We only put in barriers around volatile refs (they are expensive),
2095 // not _between_ memory refs (that would require us to track the
2096 // flavor of the previous memory refs). Requirements (2) and (3)
2097 // require some barriers before volatile stores and after volatile
2098 // loads. These nearly cover requirement (1) but miss the
2099 // volatile-store-volatile-load case. This final case is placed after
2100 // volatile-stores although it could just as well go before
2101 // volatile-loads.
2102 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2103 order_constraint) {
2104 // Helper function to insert a is-volatile test and memory barrier
2105 if (os::is_MP()) { // Not needed on single CPU
2106 __ membar(order_constraint);
2107 }
2108 }
2109
2110 void TemplateTable::resolve_cache_and_index(int byte_no,
2111 Register Rcache,
2112 Register index,
2113 size_t index_size) {
2114 const Register temp = rbx;
2115 assert_different_registers(Rcache, index, temp);
2116
2117 Label resolved;
2118 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2119 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2120 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode?
2121 __ jcc(Assembler::equal, resolved);
2122
2123 // resolve first time through
2124 address entry;
2125 switch (bytecode()) {
2126 case Bytecodes::_getstatic:
2127 case Bytecodes::_putstatic:
2128 case Bytecodes::_getfield:
2129 case Bytecodes::_putfield:
2130 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2131 break;
2132 case Bytecodes::_invokevirtual:
2133 case Bytecodes::_invokespecial:
2134 case Bytecodes::_invokestatic:
2135 case Bytecodes::_invokeinterface:
2136 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2137 break;
2138 case Bytecodes::_invokehandle:
2139 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2140 break;
2141 case Bytecodes::_invokedynamic:
2142 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2143 break;
2144 default:
2145 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2146 break;
2147 }
2148 __ movl(temp, (int) bytecode());
2149 __ call_VM(noreg, entry, temp);
2150
2151 // Update registers with resolved info
2152 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2153 __ bind(resolved);
2154 }
2155
2156 // The cache and index registers must be set before call
2157 void TemplateTable::load_field_cp_cache_entry(Register obj,
2158 Register cache,
2159 Register index,
2160 Register off,
2161 Register flags,
2162 bool is_static = false) {
2163 assert_different_registers(cache, index, flags, off);
2164
2165 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2166 // Field offset
2167 __ movptr(off, Address(cache, index, Address::times_ptr,
2168 in_bytes(cp_base_offset +
2169 ConstantPoolCacheEntry::f2_offset())));
2170 // Flags
2171 __ movl(flags, Address(cache, index, Address::times_ptr,
2172 in_bytes(cp_base_offset +
2173 ConstantPoolCacheEntry::flags_offset())));
2174
2175 // klass overwrite register
2176 if (is_static) {
2177 __ movptr(obj, Address(cache, index, Address::times_ptr,
2178 in_bytes(cp_base_offset +
2179 ConstantPoolCacheEntry::f1_offset())));
2180 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2181 __ movptr(obj, Address(obj, mirror_offset));
2182 }
2183 }
2184
2185 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2186 Register method,
2187 Register itable_index,
2188 Register flags,
2189 bool is_invokevirtual,
2190 bool is_invokevfinal, /*unused*/
2191 bool is_invokedynamic) {
2192 // setup registers
2193 const Register cache = rcx;
2194 const Register index = rdx;
2195 assert_different_registers(method, flags);
2196 assert_different_registers(method, cache, index);
2197 assert_different_registers(itable_index, flags);
2198 assert_different_registers(itable_index, cache, index);
2199 // determine constant pool cache field offsets
2200 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2201 const int method_offset = in_bytes(
2202 ConstantPoolCache::base_offset() +
2203 ((byte_no == f2_byte)
2204 ? ConstantPoolCacheEntry::f2_offset()
2205 : ConstantPoolCacheEntry::f1_offset()));
2206 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2207 ConstantPoolCacheEntry::flags_offset());
2208 // access constant pool cache fields
2209 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2210 ConstantPoolCacheEntry::f2_offset());
2211
2212 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2213 resolve_cache_and_index(byte_no, cache, index, index_size);
2214 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2215
2216 if (itable_index != noreg) {
2217 // pick up itable or appendix index from f2 also:
2218 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2219 }
2220 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2221 }
2222
2223 // Correct values of the cache and index registers are preserved.
2224 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2225 bool is_static, bool has_tos) {
2226 // do the JVMTI work here to avoid disturbing the register state below
2227 // We use c_rarg registers here because we want to use the register used in
2228 // the call to the VM
2229 if (JvmtiExport::can_post_field_access()) {
2230 // Check to see if a field access watch has been set before we
2231 // take the time to call into the VM.
2232 Label L1;
2233 assert_different_registers(cache, index, rax);
2234 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2235 __ testl(rax, rax);
2236 __ jcc(Assembler::zero, L1);
2237
2238 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2239
2240 // cache entry pointer
2241 __ addptr(c_rarg2, in_bytes(ConstantPoolCache::base_offset()));
2242 __ shll(c_rarg3, LogBytesPerWord);
2243 __ addptr(c_rarg2, c_rarg3);
2244 if (is_static) {
2245 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2246 } else {
2247 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2248 __ verify_oop(c_rarg1);
2249 }
2250 // c_rarg1: object pointer or NULL
2251 // c_rarg2: cache entry pointer
2252 // c_rarg3: jvalue object on the stack
2253 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2254 InterpreterRuntime::post_field_access),
2255 c_rarg1, c_rarg2, c_rarg3);
2256 __ get_cache_and_index_at_bcp(cache, index, 1);
2257 __ bind(L1);
2258 }
2259 }
2260
2261 void TemplateTable::pop_and_check_object(Register r) {
2262 __ pop_ptr(r);
2263 __ null_check(r); // for field access must check obj.
2264 __ verify_oop(r);
2265 }
2266
2267 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2268 transition(vtos, vtos);
2269
2270 const Register cache = rcx;
2271 const Register index = rdx;
2272 const Register obj = c_rarg3;
2273 const Register off = rbx;
2274 const Register flags = rax;
2275 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2276
2277 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2278 jvmti_post_field_access(cache, index, is_static, false);
2279 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2280
2281 if (!is_static) {
2282 // obj is on the stack
2283 pop_and_check_object(obj);
2284 }
2285
2286 const Address field(obj, off, Address::times_1);
2287
2288 Label Done, notByte, notInt, notShort, notChar,
2289 notLong, notFloat, notObj, notDouble;
2290
2291 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2292 // Make sure we don't need to mask edx after the above shift
2293 assert(btos == 0, "change code, btos != 0");
2294
2295 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2296 __ jcc(Assembler::notZero, notByte);
2297 // btos
2298 __ load_signed_byte(rax, field);
2299 __ push(btos);
2300 // Rewrite bytecode to be faster
2301 if (!is_static) {
2302 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2303 }
2304 __ jmp(Done);
2305
2306 __ bind(notByte);
2307 __ cmpl(flags, atos);
2308 __ jcc(Assembler::notEqual, notObj);
2309 // atos
2310 __ load_heap_oop(rax, field);
2311 __ push(atos);
2312 if (!is_static) {
2313 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2314 }
2315 __ jmp(Done);
2316
2317 __ bind(notObj);
2318 __ cmpl(flags, itos);
2319 __ jcc(Assembler::notEqual, notInt);
2320 // itos
2321 __ movl(rax, field);
2322 __ push(itos);
2323 // Rewrite bytecode to be faster
2324 if (!is_static) {
2325 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2326 }
2327 __ jmp(Done);
2328
2329 __ bind(notInt);
2330 __ cmpl(flags, ctos);
2331 __ jcc(Assembler::notEqual, notChar);
2332 // ctos
2333 __ load_unsigned_short(rax, field);
2334 __ push(ctos);
2335 // Rewrite bytecode to be faster
2336 if (!is_static) {
2337 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2338 }
2339 __ jmp(Done);
2340
2341 __ bind(notChar);
2342 __ cmpl(flags, stos);
2343 __ jcc(Assembler::notEqual, notShort);
2344 // stos
2345 __ load_signed_short(rax, field);
2346 __ push(stos);
2347 // Rewrite bytecode to be faster
2348 if (!is_static) {
2349 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2350 }
2351 __ jmp(Done);
2352
2353 __ bind(notShort);
2354 __ cmpl(flags, ltos);
2355 __ jcc(Assembler::notEqual, notLong);
2356 // ltos
2357 __ movq(rax, field);
2358 __ push(ltos);
2359 // Rewrite bytecode to be faster
2360 if (!is_static) {
2361 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2362 }
2363 __ jmp(Done);
2364
2365 __ bind(notLong);
2366 __ cmpl(flags, ftos);
2367 __ jcc(Assembler::notEqual, notFloat);
2368 // ftos
2369 __ movflt(xmm0, field);
2370 __ push(ftos);
2371 // Rewrite bytecode to be faster
2372 if (!is_static) {
2373 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2374 }
2375 __ jmp(Done);
2376
2377 __ bind(notFloat);
2378 #ifdef ASSERT
2379 __ cmpl(flags, dtos);
2380 __ jcc(Assembler::notEqual, notDouble);
2381 #endif
2382 // dtos
2383 __ movdbl(xmm0, field);
2384 __ push(dtos);
2385 // Rewrite bytecode to be faster
2386 if (!is_static) {
2387 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2388 }
2389 #ifdef ASSERT
2390 __ jmp(Done);
2391
2392 __ bind(notDouble);
2393 __ stop("Bad state");
2394 #endif
2395
2396 __ bind(Done);
2397 // [jk] not needed currently
2398 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2399 // Assembler::LoadStore));
2400 }
2401
2402
2403 void TemplateTable::getfield(int byte_no) {
2404 getfield_or_static(byte_no, false);
2405 }
2406
2407 void TemplateTable::getstatic(int byte_no) {
2408 getfield_or_static(byte_no, true);
2409 }
2410
2411 // The registers cache and index expected to be set before call.
2412 // The function may destroy various registers, just not the cache and index registers.
2413 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2414 transition(vtos, vtos);
2415
2416 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2417
2418 if (JvmtiExport::can_post_field_modification()) {
2419 // Check to see if a field modification watch has been set before
2420 // we take the time to call into the VM.
2421 Label L1;
2422 assert_different_registers(cache, index, rax);
2423 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2424 __ testl(rax, rax);
2425 __ jcc(Assembler::zero, L1);
2426
2427 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2428
2429 if (is_static) {
2430 // Life is simple. Null out the object pointer.
2431 __ xorl(c_rarg1, c_rarg1);
2432 } else {
2433 // Life is harder. The stack holds the value on top, followed by
2434 // the object. We don't know the size of the value, though; it
2435 // could be one or two words depending on its type. As a result,
2436 // we must find the type to determine where the object is.
2437 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2438 Address::times_8,
2439 in_bytes(cp_base_offset +
2440 ConstantPoolCacheEntry::flags_offset())));
2441 __ shrl(c_rarg3, ConstantPoolCacheEntry::tos_state_shift);
2442 // Make sure we don't need to mask rcx after the above shift
2443 ConstantPoolCacheEntry::verify_tos_state_shift();
2444 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2445 __ cmpl(c_rarg3, ltos);
2446 __ cmovptr(Assembler::equal,
2447 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2448 __ cmpl(c_rarg3, dtos);
2449 __ cmovptr(Assembler::equal,
2450 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2451 }
2452 // cache entry pointer
2453 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2454 __ shll(rscratch1, LogBytesPerWord);
2455 __ addptr(c_rarg2, rscratch1);
2456 // object (tos)
2457 __ mov(c_rarg3, rsp);
2458 // c_rarg1: object pointer set up above (NULL if static)
2459 // c_rarg2: cache entry pointer
2460 // c_rarg3: jvalue object on the stack
2461 __ call_VM(noreg,
2462 CAST_FROM_FN_PTR(address,
2463 InterpreterRuntime::post_field_modification),
2464 c_rarg1, c_rarg2, c_rarg3);
2465 __ get_cache_and_index_at_bcp(cache, index, 1);
2466 __ bind(L1);
2467 }
2468 }
2469
2470 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2471 transition(vtos, vtos);
2472
2473 const Register cache = rcx;
2474 const Register index = rdx;
2475 const Register obj = rcx;
2476 const Register off = rbx;
2477 const Register flags = rax;
2478 const Register bc = c_rarg3;
2479
2480 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2481 jvmti_post_field_mod(cache, index, is_static);
2482 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2483
2484 // [jk] not needed currently
2485 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2486 // Assembler::StoreStore));
2487
2488 Label notVolatile, Done;
2489 __ movl(rdx, flags);
2490 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2491 __ andl(rdx, 0x1);
2492
2493 // field address
2494 const Address field(obj, off, Address::times_1);
2495
2496 Label notByte, notInt, notShort, notChar,
2497 notLong, notFloat, notObj, notDouble;
2498
2499 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2500
2501 assert(btos == 0, "change code, btos != 0");
2502 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2503 __ jcc(Assembler::notZero, notByte);
2504
2505 // btos
2506 {
2507 __ pop(btos);
2508 if (!is_static) pop_and_check_object(obj);
2509 __ movb(field, rax);
2510 if (!is_static) {
2511 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
2512 }
2513 __ jmp(Done);
2514 }
2515
2516 __ bind(notByte);
2517 __ cmpl(flags, atos);
2518 __ jcc(Assembler::notEqual, notObj);
2519
2520 // atos
2521 {
2522 __ pop(atos);
2523 if (!is_static) pop_and_check_object(obj);
2524 // Store into the field
2525 do_oop_store(_masm, field, rax, _bs->kind(), false);
2526 if (!is_static) {
2527 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
2528 }
2529 __ jmp(Done);
2530 }
2531
2532 __ bind(notObj);
2533 __ cmpl(flags, itos);
2534 __ jcc(Assembler::notEqual, notInt);
2535
2536 // itos
2537 {
2538 __ pop(itos);
2539 if (!is_static) pop_and_check_object(obj);
2540 __ movl(field, rax);
2541 if (!is_static) {
2542 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
2543 }
2544 __ jmp(Done);
2545 }
2546
2547 __ bind(notInt);
2548 __ cmpl(flags, ctos);
2549 __ jcc(Assembler::notEqual, notChar);
2550
2551 // ctos
2552 {
2553 __ pop(ctos);
2554 if (!is_static) pop_and_check_object(obj);
2555 __ movw(field, rax);
2556 if (!is_static) {
2557 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
2558 }
2559 __ jmp(Done);
2560 }
2561
2562 __ bind(notChar);
2563 __ cmpl(flags, stos);
2564 __ jcc(Assembler::notEqual, notShort);
2565
2566 // stos
2567 {
2568 __ pop(stos);
2569 if (!is_static) pop_and_check_object(obj);
2570 __ movw(field, rax);
2571 if (!is_static) {
2572 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
2573 }
2574 __ jmp(Done);
2575 }
2576
2577 __ bind(notShort);
2578 __ cmpl(flags, ltos);
2579 __ jcc(Assembler::notEqual, notLong);
2580
2581 // ltos
2582 {
2583 __ pop(ltos);
2584 if (!is_static) pop_and_check_object(obj);
2585 __ movq(field, rax);
2586 if (!is_static) {
2587 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
2588 }
2589 __ jmp(Done);
2590 }
2591
2592 __ bind(notLong);
2593 __ cmpl(flags, ftos);
2594 __ jcc(Assembler::notEqual, notFloat);
2595
2596 // ftos
2597 {
2598 __ pop(ftos);
2599 if (!is_static) pop_and_check_object(obj);
2600 __ movflt(field, xmm0);
2601 if (!is_static) {
2602 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
2603 }
2604 __ jmp(Done);
2605 }
2606
2607 __ bind(notFloat);
2608 #ifdef ASSERT
2609 __ cmpl(flags, dtos);
2610 __ jcc(Assembler::notEqual, notDouble);
2611 #endif
2612
2613 // dtos
2614 {
2615 __ pop(dtos);
2616 if (!is_static) pop_and_check_object(obj);
2617 __ movdbl(field, xmm0);
2618 if (!is_static) {
2619 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
2620 }
2621 }
2622
2623 #ifdef ASSERT
2624 __ jmp(Done);
2625
2626 __ bind(notDouble);
2627 __ stop("Bad state");
2628 #endif
2629
2630 __ bind(Done);
2631
2632 // Check for volatile store
2633 __ testl(rdx, rdx);
2634 __ jcc(Assembler::zero, notVolatile);
2635 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2636 Assembler::StoreStore));
2637 __ bind(notVolatile);
2638 }
2639
2640 void TemplateTable::putfield(int byte_no) {
2641 putfield_or_static(byte_no, false);
2642 }
2643
2644 void TemplateTable::putstatic(int byte_no) {
2645 putfield_or_static(byte_no, true);
2646 }
2647
2648 void TemplateTable::jvmti_post_fast_field_mod() {
2649 if (JvmtiExport::can_post_field_modification()) {
2650 // Check to see if a field modification watch has been set before
2651 // we take the time to call into the VM.
2652 Label L2;
2653 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2654 __ testl(c_rarg3, c_rarg3);
2655 __ jcc(Assembler::zero, L2);
2656 __ pop_ptr(rbx); // copy the object pointer from tos
2657 __ verify_oop(rbx);
2658 __ push_ptr(rbx); // put the object pointer back on tos
2659 // Save tos values before call_VM() clobbers them. Since we have
2660 // to do it for every data type, we use the saved values as the
2661 // jvalue object.
2662 switch (bytecode()) { // load values into the jvalue object
2663 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
2664 case Bytecodes::_fast_bputfield: // fall through
2665 case Bytecodes::_fast_sputfield: // fall through
2666 case Bytecodes::_fast_cputfield: // fall through
2667 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
2668 case Bytecodes::_fast_dputfield: __ push_d(); break;
2669 case Bytecodes::_fast_fputfield: __ push_f(); break;
2670 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
2671
2672 default:
2673 ShouldNotReachHere();
2674 }
2675 __ mov(c_rarg3, rsp); // points to jvalue on the stack
2676 // access constant pool cache entry
2677 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2678 __ verify_oop(rbx);
2679 // rbx: object pointer copied above
2680 // c_rarg2: cache entry pointer
2681 // c_rarg3: jvalue object on the stack
2682 __ call_VM(noreg,
2683 CAST_FROM_FN_PTR(address,
2684 InterpreterRuntime::post_field_modification),
2685 rbx, c_rarg2, c_rarg3);
2686
2687 switch (bytecode()) { // restore tos values
2688 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
2689 case Bytecodes::_fast_bputfield: // fall through
2690 case Bytecodes::_fast_sputfield: // fall through
2691 case Bytecodes::_fast_cputfield: // fall through
2692 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
2693 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2694 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2695 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
2696 }
2697 __ bind(L2);
2698 }
2699 }
2700
2701 void TemplateTable::fast_storefield(TosState state) {
2702 transition(state, vtos);
2703
2704 ByteSize base = ConstantPoolCache::base_offset();
2705
2706 jvmti_post_fast_field_mod();
2707
2708 // access constant pool cache
2709 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2710
2711 // test for volatile with rdx
2712 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2713 in_bytes(base +
2714 ConstantPoolCacheEntry::flags_offset())));
2715
2716 // replace index with field offset from cache entry
2717 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2718 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2719
2720 // [jk] not needed currently
2721 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2722 // Assembler::StoreStore));
2723
2724 Label notVolatile;
2725 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2726 __ andl(rdx, 0x1);
2727
2728 // Get object from stack
2729 pop_and_check_object(rcx);
2730
2731 // field address
2732 const Address field(rcx, rbx, Address::times_1);
2733
2734 // access field
2735 switch (bytecode()) {
2736 case Bytecodes::_fast_aputfield:
2737 do_oop_store(_masm, field, rax, _bs->kind(), false);
2738 break;
2739 case Bytecodes::_fast_lputfield:
2740 __ movq(field, rax);
2741 break;
2742 case Bytecodes::_fast_iputfield:
2743 __ movl(field, rax);
2744 break;
2745 case Bytecodes::_fast_bputfield:
2746 __ movb(field, rax);
2747 break;
2748 case Bytecodes::_fast_sputfield:
2749 // fall through
2750 case Bytecodes::_fast_cputfield:
2751 __ movw(field, rax);
2752 break;
2753 case Bytecodes::_fast_fputfield:
2754 __ movflt(field, xmm0);
2755 break;
2756 case Bytecodes::_fast_dputfield:
2757 __ movdbl(field, xmm0);
2758 break;
2759 default:
2760 ShouldNotReachHere();
2761 }
2762
2763 // Check for volatile store
2764 __ testl(rdx, rdx);
2765 __ jcc(Assembler::zero, notVolatile);
2766 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2767 Assembler::StoreStore));
2768 __ bind(notVolatile);
2769 }
2770
2771
2772 void TemplateTable::fast_accessfield(TosState state) {
2773 transition(atos, state);
2774
2775 // Do the JVMTI work here to avoid disturbing the register state below
2776 if (JvmtiExport::can_post_field_access()) {
2777 // Check to see if a field access watch has been set before we
2778 // take the time to call into the VM.
2779 Label L1;
2780 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2781 __ testl(rcx, rcx);
2782 __ jcc(Assembler::zero, L1);
2783 // access constant pool cache entry
2784 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2785 __ verify_oop(rax);
2786 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2787 __ mov(c_rarg1, rax);
2788 // c_rarg1: object pointer copied above
2789 // c_rarg2: cache entry pointer
2790 __ call_VM(noreg,
2791 CAST_FROM_FN_PTR(address,
2792 InterpreterRuntime::post_field_access),
2793 c_rarg1, c_rarg2);
2794 __ pop_ptr(rax); // restore object pointer
2795 __ bind(L1);
2796 }
2797
2798 // access constant pool cache
2799 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2800 // replace index with field offset from cache entry
2801 // [jk] not needed currently
2802 // if (os::is_MP()) {
2803 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2804 // in_bytes(ConstantPoolCache::base_offset() +
2805 // ConstantPoolCacheEntry::flags_offset())));
2806 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2807 // __ andl(rdx, 0x1);
2808 // }
2809 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2810 in_bytes(ConstantPoolCache::base_offset() +
2811 ConstantPoolCacheEntry::f2_offset())));
2812
2813 // rax: object
2814 __ verify_oop(rax);
2815 __ null_check(rax);
2816 Address field(rax, rbx, Address::times_1);
2817
2818 // access field
2819 switch (bytecode()) {
2820 case Bytecodes::_fast_agetfield:
2821 __ load_heap_oop(rax, field);
2822 __ verify_oop(rax);
2823 break;
2824 case Bytecodes::_fast_lgetfield:
2825 __ movq(rax, field);
2826 break;
2827 case Bytecodes::_fast_igetfield:
2828 __ movl(rax, field);
2829 break;
2830 case Bytecodes::_fast_bgetfield:
2831 __ movsbl(rax, field);
2832 break;
2833 case Bytecodes::_fast_sgetfield:
2834 __ load_signed_short(rax, field);
2835 break;
2836 case Bytecodes::_fast_cgetfield:
2837 __ load_unsigned_short(rax, field);
2838 break;
2839 case Bytecodes::_fast_fgetfield:
2840 __ movflt(xmm0, field);
2841 break;
2842 case Bytecodes::_fast_dgetfield:
2843 __ movdbl(xmm0, field);
2844 break;
2845 default:
2846 ShouldNotReachHere();
2847 }
2848 // [jk] not needed currently
2849 // if (os::is_MP()) {
2850 // Label notVolatile;
2851 // __ testl(rdx, rdx);
2852 // __ jcc(Assembler::zero, notVolatile);
2853 // __ membar(Assembler::LoadLoad);
2854 // __ bind(notVolatile);
2855 //};
2856 }
2857
2858 void TemplateTable::fast_xaccess(TosState state) {
2859 transition(vtos, state);
2860
2861 // get receiver
2862 __ movptr(rax, aaddress(0));
2863 // access constant pool cache
2864 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2865 __ movptr(rbx,
2866 Address(rcx, rdx, Address::times_8,
2867 in_bytes(ConstantPoolCache::base_offset() +
2868 ConstantPoolCacheEntry::f2_offset())));
2869 // make sure exception is reported in correct bcp range (getfield is
2870 // next instruction)
2871 __ increment(r13);
2872 __ null_check(rax);
2873 switch (state) {
2874 case itos:
2875 __ movl(rax, Address(rax, rbx, Address::times_1));
2876 break;
2877 case atos:
2878 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2879 __ verify_oop(rax);
2880 break;
2881 case ftos:
2882 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2883 break;
2884 default:
2885 ShouldNotReachHere();
2886 }
2887
2888 // [jk] not needed currently
2889 // if (os::is_MP()) {
2890 // Label notVolatile;
2891 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2892 // in_bytes(ConstantPoolCache::base_offset() +
2893 // ConstantPoolCacheEntry::flags_offset())));
2894 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2895 // __ testl(rdx, 0x1);
2896 // __ jcc(Assembler::zero, notVolatile);
2897 // __ membar(Assembler::LoadLoad);
2898 // __ bind(notVolatile);
2899 // }
2900
2901 __ decrement(r13);
2902 }
2903
2904
2905
2906 //-----------------------------------------------------------------------------
2907 // Calls
2908
2909 void TemplateTable::count_calls(Register method, Register temp) {
2910 // implemented elsewhere
2911 ShouldNotReachHere();
2912 }
2913
2914 void TemplateTable::prepare_invoke(int byte_no,
2915 Register method, // linked method (or i-klass)
2916 Register index, // itable index, MethodType, etc.
2917 Register recv, // if caller wants to see it
2918 Register flags // if caller wants to test it
2919 ) {
2920 // determine flags
2921 const Bytecodes::Code code = bytecode();
2922 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2923 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2924 const bool is_invokehandle = code == Bytecodes::_invokehandle;
2925 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2926 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2927 const bool load_receiver = (recv != noreg);
2928 const bool save_flags = (flags != noreg);
2929 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2930 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
2931 assert(flags == noreg || flags == rdx, "");
2932 assert(recv == noreg || recv == rcx, "");
2933
2934 // setup registers & access constant pool cache
2935 if (recv == noreg) recv = rcx;
2936 if (flags == noreg) flags = rdx;
2937 assert_different_registers(method, index, recv, flags);
2938
2939 // save 'interpreter return address'
2940 __ save_bcp();
2941
2942 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2943
2944 // maybe push appendix to arguments (just before return address)
2945 if (is_invokedynamic || is_invokehandle) {
2946 Label L_no_push;
2947 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
2948 __ jcc(Assembler::zero, L_no_push);
2949 // Push the appendix as a trailing parameter.
2950 // This must be done before we get the receiver,
2951 // since the parameter_size includes it.
2952 __ push(rbx);
2953 __ mov(rbx, index);
2954 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2955 __ load_resolved_reference_at_index(index, rbx);
2956 __ pop(rbx);
2957 __ push(index); // push appendix (MethodType, CallSite, etc.)
2958 __ bind(L_no_push);
2959 }
2960
2961 // load receiver if needed (after appendix is pushed so parameter size is correct)
2962 // Note: no return address pushed yet
2963 if (load_receiver) {
2964 __ movl(recv, flags);
2965 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
2966 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
2967 const int receiver_is_at_end = -1; // back off one slot to get receiver
2968 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
2969 __ movptr(recv, recv_addr);
2970 __ verify_oop(recv);
2971 }
2972
2973 if (save_flags) {
2974 __ movl(r13, flags);
2975 }
2976
2977 // compute return type
2978 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2979 // Make sure we don't need to mask flags after the above shift
2980 ConstantPoolCacheEntry::verify_tos_state_shift();
2981 // load return address
2982 {
2983 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2984 ExternalAddress table(table_addr);
2985 __ lea(rscratch1, table);
2986 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2987 }
2988
2989 // push return address
2990 __ push(flags);
2991
2992 // Restore flags value from the constant pool cache, and restore rsi
2993 // for later null checks. r13 is the bytecode pointer
2994 if (save_flags) {
2995 __ movl(flags, r13);
2996 __ restore_bcp();
2997 }
2998 }
2999
3000
3001 void TemplateTable::invokevirtual_helper(Register index,
3002 Register recv,
3003 Register flags) {
3004 // Uses temporary registers rax, rdx
3005 assert_different_registers(index, recv, rax, rdx);
3006 assert(index == rbx, "");
3007 assert(recv == rcx, "");
3008
3009 // Test for an invoke of a final method
3010 Label notFinal;
3011 __ movl(rax, flags);
3012 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3013 __ jcc(Assembler::zero, notFinal);
3014
3015 const Register method = index; // method must be rbx
3016 assert(method == rbx,
3017 "Method* must be rbx for interpreter calling convention");
3018
3019 // do the call - the index is actually the method to call
3020 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3021
3022 // It's final, need a null check here!
3023 __ null_check(recv);
3024
3025 // profile this call
3026 __ profile_final_call(rax);
3027 __ profile_arguments_type(rax, method, r13, true);
3028
3029 __ jump_from_interpreted(method, rax);
3030
3031 __ bind(notFinal);
3032
3033 // get receiver klass
3034 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3035 __ load_klass(rax, recv);
3036
3037 // profile this call
3038 __ profile_virtual_call(rax, r14, rdx);
3039
3040 // get target Method* & entry point
3041 __ lookup_virtual_method(rax, index, method);
3042 __ profile_arguments_type(rdx, method, r13, true);
3043 __ jump_from_interpreted(method, rdx);
3044 }
3045
3046
3047 void TemplateTable::invokevirtual(int byte_no) {
3048 transition(vtos, vtos);
3049 assert(byte_no == f2_byte, "use this argument");
3050 prepare_invoke(byte_no,
3051 rbx, // method or vtable index
3052 noreg, // unused itable index
3053 rcx, rdx); // recv, flags
3054
3055 // rbx: index
3056 // rcx: receiver
3057 // rdx: flags
3058
3059 invokevirtual_helper(rbx, rcx, rdx);
3060 }
3061
3062
3063 void TemplateTable::invokespecial(int byte_no) {
3064 transition(vtos, vtos);
3065 assert(byte_no == f1_byte, "use this argument");
3066 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3067 rcx); // get receiver also for null check
3068 __ verify_oop(rcx);
3069 __ null_check(rcx);
3070 // do the call
3071 __ profile_call(rax);
3072 __ profile_arguments_type(rax, rbx, r13, false);
3073 __ jump_from_interpreted(rbx, rax);
3074 }
3075
3076
3077 void TemplateTable::invokestatic(int byte_no) {
3078 transition(vtos, vtos);
3079 assert(byte_no == f1_byte, "use this argument");
3080 prepare_invoke(byte_no, rbx); // get f1 Method*
3081 // do the call
3082 __ profile_call(rax);
3083 __ profile_arguments_type(rax, rbx, r13, false);
3084 __ jump_from_interpreted(rbx, rax);
3085 }
3086
3087 void TemplateTable::fast_invokevfinal(int byte_no) {
3088 transition(vtos, vtos);
3089 assert(byte_no == f2_byte, "use this argument");
3090 __ stop("fast_invokevfinal not used on amd64");
3091 }
3092
3093 void TemplateTable::invokeinterface(int byte_no) {
3094 transition(vtos, vtos);
3095 assert(byte_no == f1_byte, "use this argument");
3096 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3097 rcx, rdx); // recv, flags
3098
3099 // rax: interface klass (from f1)
3100 // rbx: itable index (from f2)
3101 // rcx: receiver
3102 // rdx: flags
3103
3104 // Special case of invokeinterface called for virtual method of
3105 // java.lang.Object. See cpCacheOop.cpp for details.
3106 // This code isn't produced by javac, but could be produced by
3107 // another compliant java compiler.
3108 Label notMethod;
3109 __ movl(r14, rdx);
3110 __ andl(r14, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3111 __ jcc(Assembler::zero, notMethod);
3112
3113 invokevirtual_helper(rbx, rcx, rdx);
3114 __ bind(notMethod);
3115
3116 // Get receiver klass into rdx - also a null check
3117 __ restore_locals(); // restore r14
3118 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3119 __ load_klass(rdx, rcx);
3120
3121 // profile this call
3122 __ profile_virtual_call(rdx, r13, r14);
3123
3124 Label no_such_interface, no_such_method;
3125
3126 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3127 rdx, rax, rbx,
3128 // outputs: method, scan temp. reg
3129 rbx, r13,
3130 no_such_interface);
3131
3132 // rbx: Method* to call
3133 // rcx: receiver
3134 // Check for abstract method error
3135 // Note: This should be done more efficiently via a throw_abstract_method_error
3136 // interpreter entry point and a conditional jump to it in case of a null
3137 // method.
3138 __ testptr(rbx, rbx);
3139 __ jcc(Assembler::zero, no_such_method);
3140
3141 __ profile_arguments_type(rdx, rbx, r13, true);
3142
3143 // do the call
3144 // rcx: receiver
3145 // rbx,: Method*
3146 __ jump_from_interpreted(rbx, rdx);
3147 __ should_not_reach_here();
3148
3149 // exception handling code follows...
3150 // note: must restore interpreter registers to canonical
3151 // state for exception handling to work correctly!
3152
3153 __ bind(no_such_method);
3154 // throw exception
3155 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3156 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3157 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3158 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3159 // the call_VM checks for exception, so we should never return here.
3160 __ should_not_reach_here();
3161
3162 __ bind(no_such_interface);
3163 // throw exception
3164 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3165 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3166 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3167 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3168 InterpreterRuntime::throw_IncompatibleClassChangeError));
3169 // the call_VM checks for exception, so we should never return here.
3170 __ should_not_reach_here();
3171 }
3172
3173
3174 void TemplateTable::invokehandle(int byte_no) {
3175 transition(vtos, vtos);
3176 assert(byte_no == f1_byte, "use this argument");
3177 const Register rbx_method = rbx;
3178 const Register rax_mtype = rax;
3179 const Register rcx_recv = rcx;
3180 const Register rdx_flags = rdx;
3181
3182 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3183 __ verify_method_ptr(rbx_method);
3184 __ verify_oop(rcx_recv);
3185 __ null_check(rcx_recv);
3186
3187 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3188 // rbx: MH.invokeExact_MT method (from f2)
3189
3190 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3191
3192 // FIXME: profile the LambdaForm also
3193 __ profile_final_call(rax);
3194 __ profile_arguments_type(rdx, rbx_method, r13, true);
3195
3196 __ jump_from_interpreted(rbx_method, rdx);
3197 }
3198
3199
3200 void TemplateTable::invokedynamic(int byte_no) {
3201 transition(vtos, vtos);
3202 assert(byte_no == f1_byte, "use this argument");
3203
3204 const Register rbx_method = rbx;
3205 const Register rax_callsite = rax;
3206
3207 prepare_invoke(byte_no, rbx_method, rax_callsite);
3208
3209 // rax: CallSite object (from cpool->resolved_references[f1])
3210 // rbx: MH.linkToCallSite method (from f2)
3211
3212 // Note: rax_callsite is already pushed by prepare_invoke
3213
3214 // %%% should make a type profile for any invokedynamic that takes a ref argument
3215 // profile this call
3216 __ profile_call(r13);
3217 __ profile_arguments_type(rdx, rbx_method, r13, false);
3218
3219 __ verify_oop(rax_callsite);
3220
3221 __ jump_from_interpreted(rbx_method, rdx);
3222 }
3223
3224
3225 //-----------------------------------------------------------------------------
3226 // Allocation
3227
3228 void TemplateTable::_new() {
3229 transition(vtos, atos);
3230 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3231 Label slow_case;
3232 Label done;
3233 Label initialize_header;
3234 Label initialize_object; // including clearing the fields
3235 Label allocate_shared;
3236
3237 __ get_cpool_and_tags(rsi, rax);
3238 // Make sure the class we're about to instantiate has been resolved.
3239 // This is done before loading InstanceKlass to be consistent with the order
3240 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3241 const int tags_offset = Array<u1>::base_offset_in_bytes();
3242 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3243 JVM_CONSTANT_Class);
3244 __ jcc(Assembler::notEqual, slow_case);
3245
3246 // get InstanceKlass
3247 __ movptr(rsi, Address(rsi, rdx,
3248 Address::times_8, sizeof(ConstantPool)));
3249
3250 // make sure klass is initialized & doesn't have finalizer
3251 // make sure klass is fully initialized
3252 __ cmpb(Address(rsi,
3253 InstanceKlass::init_state_offset()),
3254 InstanceKlass::fully_initialized);
3255 __ jcc(Assembler::notEqual, slow_case);
3256
3257 // get instance_size in InstanceKlass (scaled to a count of bytes)
3258 __ movl(rdx,
3259 Address(rsi,
3260 Klass::layout_helper_offset()));
3261 // test to see if it has a finalizer or is malformed in some way
3262 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3263 __ jcc(Assembler::notZero, slow_case);
3264
3265 // Allocate the instance
3266 // 1) Try to allocate in the TLAB
3267 // 2) if fail and the object is large allocate in the shared Eden
3268 // 3) if the above fails (or is not applicable), go to a slow case
3269 // (creates a new TLAB, etc.)
3270
3271 const bool allow_shared_alloc =
3272 Universe::heap()->supports_inline_contig_alloc();
3273
3274 if (UseTLAB) {
3275 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3276 __ lea(rbx, Address(rax, rdx, Address::times_1));
3277 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3278 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3279 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3280 if (ZeroTLAB) {
3281 // the fields have been already cleared
3282 __ jmp(initialize_header);
3283 } else {
3284 // initialize both the header and fields
3285 __ jmp(initialize_object);
3286 }
3287 }
3288
3289 // Allocation in the shared Eden, if allowed.
3290 //
3291 // rdx: instance size in bytes
3292 if (allow_shared_alloc) {
3293 __ bind(allocate_shared);
3294
3295 ExternalAddress top((address)Universe::heap()->top_addr());
3296 ExternalAddress end((address)Universe::heap()->end_addr());
3297
3298 const Register RtopAddr = rscratch1;
3299 const Register RendAddr = rscratch2;
3300
3301 __ lea(RtopAddr, top);
3302 __ lea(RendAddr, end);
3303 __ movptr(rax, Address(RtopAddr, 0));
3304
3305 // For retries rax gets set by cmpxchgq
3306 Label retry;
3307 __ bind(retry);
3308 __ lea(rbx, Address(rax, rdx, Address::times_1));
3309 __ cmpptr(rbx, Address(RendAddr, 0));
3310 __ jcc(Assembler::above, slow_case);
3311
3312 // Compare rax with the top addr, and if still equal, store the new
3313 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3314 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3315 //
3316 // rax: object begin
3317 // rbx: object end
3318 // rdx: instance size in bytes
3319 if (os::is_MP()) {
3320 __ lock();
3321 }
3322 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3323
3324 // if someone beat us on the allocation, try again, otherwise continue
3325 __ jcc(Assembler::notEqual, retry);
3326
3327 __ incr_allocated_bytes(r15_thread, rdx, 0);
3328 }
3329
3330 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3331 // The object is initialized before the header. If the object size is
3332 // zero, go directly to the header initialization.
3333 __ bind(initialize_object);
3334 __ decrementl(rdx, sizeof(oopDesc));
3335 __ jcc(Assembler::zero, initialize_header);
3336
3337 // Initialize object fields
3338 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3339 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3340 {
3341 Label loop;
3342 __ bind(loop);
3343 __ movq(Address(rax, rdx, Address::times_8,
3344 sizeof(oopDesc) - oopSize),
3345 rcx);
3346 __ decrementl(rdx);
3347 __ jcc(Assembler::notZero, loop);
3348 }
3349
3350 // initialize object header only.
3351 __ bind(initialize_header);
3352 if (UseBiasedLocking) {
3353 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset()));
3354 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3355 } else {
3356 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3357 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3358 }
3359 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3360 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3361 __ store_klass(rax, rsi); // store klass last
3362
3363 {
3364 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3365 // Trigger dtrace event for fastpath
3366 __ push(atos); // save the return value
3367 __ call_VM_leaf(
3368 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3369 __ pop(atos); // restore the return value
3370
3371 }
3372 __ jmp(done);
3373 }
3374
3375
3376 // slow case
3377 __ bind(slow_case);
3378 __ get_constant_pool(c_rarg1);
3379 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3380 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3381 __ verify_oop(rax);
3382
3383 // continue
3384 __ bind(done);
3385 }
3386
3387 void TemplateTable::newarray() {
3388 transition(itos, atos);
3389 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3390 __ movl(c_rarg2, rax);
3391 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3392 c_rarg1, c_rarg2);
3393 }
3394
3395 void TemplateTable::anewarray() {
3396 transition(itos, atos);
3397 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3398 __ get_constant_pool(c_rarg1);
3399 __ movl(c_rarg3, rax);
3400 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3401 c_rarg1, c_rarg2, c_rarg3);
3402 }
3403
3404 void TemplateTable::arraylength() {
3405 transition(atos, itos);
3406 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3407 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3408 }
3409
3410 void TemplateTable::checkcast() {
3411 transition(atos, atos);
3412 Label done, is_null, ok_is_subtype, quicked, resolved;
3413 __ testptr(rax, rax); // object is in rax
3414 __ jcc(Assembler::zero, is_null);
3415
3416 // Get cpool & tags index
3417 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3418 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3419 // See if bytecode has already been quicked
3420 __ cmpb(Address(rdx, rbx,
3421 Address::times_1,
3422 Array<u1>::base_offset_in_bytes()),
3423 JVM_CONSTANT_Class);
3424 __ jcc(Assembler::equal, quicked);
3425 __ push(atos); // save receiver for result, and for GC
3426 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3427 // vm_result_2 has metadata result
3428 __ get_vm_result_2(rax, r15_thread);
3429 __ pop_ptr(rdx); // restore receiver
3430 __ jmpb(resolved);
3431
3432 // Get superklass in rax and subklass in rbx
3433 __ bind(quicked);
3434 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3435 __ movptr(rax, Address(rcx, rbx,
3436 Address::times_8, sizeof(ConstantPool)));
3437
3438 __ bind(resolved);
3439 __ load_klass(rbx, rdx);
3440
3441 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3442 // Superklass in rax. Subklass in rbx.
3443 __ gen_subtype_check(rbx, ok_is_subtype);
3444
3445 // Come here on failure
3446 __ push_ptr(rdx);
3447 // object is at TOS
3448 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3449
3450 // Come here on success
3451 __ bind(ok_is_subtype);
3452 __ mov(rax, rdx); // Restore object in rdx
3453
3454 // Collect counts on whether this check-cast sees NULLs a lot or not.
3455 if (ProfileInterpreter) {
3456 __ jmp(done);
3457 __ bind(is_null);
3458 __ profile_null_seen(rcx);
3459 } else {
3460 __ bind(is_null); // same as 'done'
3461 }
3462 __ bind(done);
3463 }
3464
3465 void TemplateTable::instanceof() {
3466 transition(atos, itos);
3467 Label done, is_null, ok_is_subtype, quicked, resolved;
3468 __ testptr(rax, rax);
3469 __ jcc(Assembler::zero, is_null);
3470
3471 // Get cpool & tags index
3472 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3473 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3474 // See if bytecode has already been quicked
3475 __ cmpb(Address(rdx, rbx,
3476 Address::times_1,
3477 Array<u1>::base_offset_in_bytes()),
3478 JVM_CONSTANT_Class);
3479 __ jcc(Assembler::equal, quicked);
3480
3481 __ push(atos); // save receiver for result, and for GC
3482 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3483 // vm_result_2 has metadata result
3484 __ get_vm_result_2(rax, r15_thread);
3485 __ pop_ptr(rdx); // restore receiver
3486 __ verify_oop(rdx);
3487 __ load_klass(rdx, rdx);
3488 __ jmpb(resolved);
3489
3490 // Get superklass in rax and subklass in rdx
3491 __ bind(quicked);
3492 __ load_klass(rdx, rax);
3493 __ movptr(rax, Address(rcx, rbx,
3494 Address::times_8, sizeof(ConstantPool)));
3495
3496 __ bind(resolved);
3497
3498 // Generate subtype check. Blows rcx, rdi
3499 // Superklass in rax. Subklass in rdx.
3500 __ gen_subtype_check(rdx, ok_is_subtype);
3501
3502 // Come here on failure
3503 __ xorl(rax, rax);
3504 __ jmpb(done);
3505 // Come here on success
3506 __ bind(ok_is_subtype);
3507 __ movl(rax, 1);
3508
3509 // Collect counts on whether this test sees NULLs a lot or not.
3510 if (ProfileInterpreter) {
3511 __ jmp(done);
3512 __ bind(is_null);
3513 __ profile_null_seen(rcx);
3514 } else {
3515 __ bind(is_null); // same as 'done'
3516 }
3517 __ bind(done);
3518 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3519 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3520 }
3521
3522 //-----------------------------------------------------------------------------
3523 // Breakpoints
3524 void TemplateTable::_breakpoint() {
3525 // Note: We get here even if we are single stepping..
3526 // jbug inists on setting breakpoints at every bytecode
3527 // even if we are in single step mode.
3528
3529 transition(vtos, vtos);
3530
3531 // get the unpatched byte code
3532 __ get_method(c_rarg1);
3533 __ call_VM(noreg,
3534 CAST_FROM_FN_PTR(address,
3535 InterpreterRuntime::get_original_bytecode_at),
3536 c_rarg1, r13);
3537 __ mov(rbx, rax);
3538
3539 // post the breakpoint event
3540 __ get_method(c_rarg1);
3541 __ call_VM(noreg,
3542 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3543 c_rarg1, r13);
3544
3545 // complete the execution of original bytecode
3546 __ dispatch_only_normal(vtos);
3547 }
3548
3549 //-----------------------------------------------------------------------------
3550 // Exceptions
3551
3552 void TemplateTable::athrow() {
3553 transition(atos, vtos);
3554 __ null_check(rax);
3555 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3556 }
3557
3558 //-----------------------------------------------------------------------------
3559 // Synchronization
3560 //
3561 // Note: monitorenter & exit are symmetric routines; which is reflected
3562 // in the assembly code structure as well
3563 //
3564 // Stack layout:
3565 //
3566 // [expressions ] <--- rsp = expression stack top
3567 // ..
3568 // [expressions ]
3569 // [monitor entry] <--- monitor block top = expression stack bot
3570 // ..
3571 // [monitor entry]
3572 // [frame data ] <--- monitor block bot
3573 // ...
3574 // [saved rbp ] <--- rbp
3575 void TemplateTable::monitorenter() {
3576 transition(atos, vtos);
3577
3578 // check for NULL object
3579 __ null_check(rax);
3580
3581 const Address monitor_block_top(
3582 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3583 const Address monitor_block_bot(
3584 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3585 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3586
3587 Label allocated;
3588
3589 // initialize entry pointer
3590 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3591
3592 // find a free slot in the monitor block (result in c_rarg1)
3593 {
3594 Label entry, loop, exit;
3595 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3596 // starting with top-most entry
3597 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3598 // of monitor block
3599 __ jmpb(entry);
3600
3601 __ bind(loop);
3602 // check if current entry is used
3603 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3604 // if not used then remember entry in c_rarg1
3605 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3606 // check if current entry is for same object
3607 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3608 // if same object then stop searching
3609 __ jccb(Assembler::equal, exit);
3610 // otherwise advance to next entry
3611 __ addptr(c_rarg3, entry_size);
3612 __ bind(entry);
3613 // check if bottom reached
3614 __ cmpptr(c_rarg3, c_rarg2);
3615 // if not at bottom then check this entry
3616 __ jcc(Assembler::notEqual, loop);
3617 __ bind(exit);
3618 }
3619
3620 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3621 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3622
3623 // allocate one if there's no free slot
3624 {
3625 Label entry, loop;
3626 // 1. compute new pointers // rsp: old expression stack top
3627 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3628 __ subptr(rsp, entry_size); // move expression stack top
3629 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3630 __ mov(c_rarg3, rsp); // set start value for copy loop
3631 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3632 __ jmp(entry);
3633 // 2. move expression stack contents
3634 __ bind(loop);
3635 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3636 // word from old location
3637 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3638 __ addptr(c_rarg3, wordSize); // advance to next word
3639 __ bind(entry);
3640 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3641 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3642 // copy next word
3643 }
3644
3645 // call run-time routine
3646 // c_rarg1: points to monitor entry
3647 __ bind(allocated);
3648
3649 // Increment bcp to point to the next bytecode, so exception
3650 // handling for async. exceptions work correctly.
3651 // The object has already been poped from the stack, so the
3652 // expression stack looks correct.
3653 __ increment(r13);
3654
3655 // store object
3656 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3657 __ lock_object(c_rarg1);
3658
3659 // check to make sure this monitor doesn't cause stack overflow after locking
3660 __ save_bcp(); // in case of exception
3661 __ generate_stack_overflow_check(0);
3662
3663 // The bcp has already been incremented. Just need to dispatch to
3664 // next instruction.
3665 __ dispatch_next(vtos);
3666 }
3667
3668
3669 void TemplateTable::monitorexit() {
3670 transition(atos, vtos);
3671
3672 // check for NULL object
3673 __ null_check(rax);
3674
3675 const Address monitor_block_top(
3676 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3677 const Address monitor_block_bot(
3678 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3679 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3680
3681 Label found;
3682
3683 // find matching slot
3684 {
3685 Label entry, loop;
3686 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3687 // starting with top-most entry
3688 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3689 // of monitor block
3690 __ jmpb(entry);
3691
3692 __ bind(loop);
3693 // check if current entry is for same object
3694 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3695 // if same object then stop searching
3696 __ jcc(Assembler::equal, found);
3697 // otherwise advance to next entry
3698 __ addptr(c_rarg1, entry_size);
3699 __ bind(entry);
3700 // check if bottom reached
3701 __ cmpptr(c_rarg1, c_rarg2);
3702 // if not at bottom then check this entry
3703 __ jcc(Assembler::notEqual, loop);
3704 }
3705
3706 // error handling. Unlocking was not block-structured
3707 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3708 InterpreterRuntime::throw_illegal_monitor_state_exception));
3709 __ should_not_reach_here();
3710
3711 // call run-time routine
3712 // rsi: points to monitor entry
3713 __ bind(found);
3714 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3715 __ unlock_object(c_rarg1);
3716 __ pop_ptr(rax); // discard object
3717 }
3718
3719
3720 // Wide instructions
3721 void TemplateTable::wide() {
3722 transition(vtos, vtos);
3723 __ load_unsigned_byte(rbx, at_bcp(1));
3724 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3725 __ jmp(Address(rscratch1, rbx, Address::times_8));
3726 // Note: the r13 increment step is part of the individual wide
3727 // bytecode implementations
3728 }
3729
3730
3731 // Multi arrays
3732 void TemplateTable::multianewarray() {
3733 transition(vtos, atos);
3734 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3735 // last dim is on top of stack; we want address of first one:
3736 // first_addr = last_addr + (ndims - 1) * wordSize
3737 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3738 call_VM(rax,
3739 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3740 c_rarg1);
3741 __ load_unsigned_byte(rbx, at_bcp(3));
3742 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3743 }
3744 #endif // !CC_INTERP